embodichain.lab.sim.atomic_actions.primitives

Built-in atomic action primitive implementations.

Classes:

AssembleGoal	Place a held assemble object onto a base object at a relative pose.
AxisAlign	Grasp an object and align its local axis to a world axis.
AxisAlignGoal	Object whose local axis should be aligned after an antipodal grasp.
AxisAlignOptions	Per-invocation grasp-and-axis-alignment behavior.
CoordinatedPickGoal	Object-centric target for picking and moving one object with two hands.
CoordinatedPickment	Pick and move a single object pinched by two hands.
CoordinatedPickmentOptions	Per-invocation coordinated pickup behavior.
CoordinatedPlacement	Coordinate two held objects: support object below, placing object above.
CoordinatedPlacementGoal	Object-centric target for dual-arm coordinated placement.
CoordinatedPlacementOptions	Per-invocation coordinated placement behavior.
EndEffectorPoseGoal	End-effector pose goal with optional batched intermediate waypoints.
GraspGoal	Pickup target with an affordance-selected or supplied grasp pose.
HandOver	Pick an object with the nearer arm, hand it over, and place it.
HandOverGoal	Object to pick and hand over, plus its final object pose.
HandOverOptions	Per-invocation pick-up, handover, and placement behavior.
HeldObjectPoseGoal	Desired pose for the object held by this action's control part.
JointPositionGoal	Explicit or named joint-space goal for a bound robot resource.
MoveEndEffector	Plan a free-space move for a bound manipulator.
MoveEndEffectorOptions	Per-invocation behavior for MoveEndEffector.
MoveHeldObject	Move the held object to the exact target object pose with a closed hand.
MoveHeldObjectOptions	Per-invocation held-object transport behavior.
MoveJoints	Plan joint motion from the observed state to one or more waypoints.
MoveJointsOptions	Per-invocation behavior for MoveJoints.
OpenDoor	Approach, grasp, rotate a door about its hinge, release, and retract.
OpenDoorGoal	Door handle and desired absolute opening state.
OpenDoorOptions	Per-invocation approach, interpolation, release, and retract behavior.
PickUp	Approach a grasp pose, close the gripper, lift.
PickUpOptions	Per-invocation pickup behavior.
Place	Lower the held object to a place pose, open the gripper, retract.
PlaceGoal	End-effector release-pose target used by Place.
PlaceOptions	Per-invocation placement behavior.
Pour	Rotate and return an exclusively held object about its internal axis.
PourGoal	Rotate the object currently held by the bound manipulator.
PourOptions	Per-invocation pouring behavior.
Press	Open-loop motion primitive that approaches, presses, and retracts.
PressGoal	Target object described by a press affordance.
PressOptions	Per-invocation pressing behavior.
PushObject	Close the end effector, contact a rigid object, and push it in-plane.
PushObjectGoal	Push one rigid object toward a target pose on the target support plane.
PushObjectOptions	Per-invocation planar pushing behavior.
PushObjectToolCalibration	End-effector calibration selected by a bound motion control part.
Slide	Open-loop approach, grasp, and axis-constrained sliding motion.
SlideGoal	Translating articulation link described by a slide affordance.
SlideOptions	Per-invocation sliding behavior for a translating articulation link.
Twist	Open-loop approach, grasp, twist, release, and retract motion.
TwistGoal	Target object described by a twist affordance.
TwistOptions	Per-invocation twisting behavior.

Data:

BUILTIN_ACTION_TYPES

Built-in action implementations instantiated once per action engine.

class embodichain.lab.sim.atomic_actions.primitives.AssembleGoal

Place a held assemble object onto a base object at a relative pose.

The base object pose is a late-bound SceneEntityPose. The assemble object’s target pose is base_pose @ assemble_to_base_pose. The held-object transform (object_to_eef) is read from PlanningContext for the place control part, which a prior PickUp populates.

Methods:

__init__(affordance, base_pose)

Attributes:

affordance	Assembly affordance anchoring the assemble object to the base object.
base_pose	Late-bound base-object pose used for snapshot-consistent planning.

__init__(affordance, base_pose)

affordance: AssembleAffordance

Assembly affordance anchoring the assemble object to the base object.

base_pose: SceneEntityPose

Late-bound base-object pose used for snapshot-consistent planning.

class embodichain.lab.sim.atomic_actions.primitives.AxisAlign

Grasp an object and align its local axis to a world axis.

Classes:

GoalType	alias of AxisAlignGoal
OptionsType	alias of AxisAlignOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of AxisAlignGoal

OptionsType

alias of AxisAlignOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'axis_align'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.AxisAlignGoal

Object whose local axis should be aligned after an antipodal grasp.

Methods:

__init__(semantics[, grasp_xpos])

Attributes:

grasp_xpos

Optional explicit end-effector grasp pose; omitted poses are sampled.

__init__(semantics, grasp_xpos=None)

grasp_xpos: Tensor | SceneEntityPose | None

Optional explicit end-effector grasp pose; omitted poses are sampled.

class embodichain.lab.sim.atomic_actions.primitives.AxisAlignOptions

Per-invocation grasp-and-axis-alignment behavior.

Methods:

__init__([hand_interp_steps, ...])

Attributes:

target_axis

Desired world-frame axis, shape (3,) or (B, 3).

__init__(hand_interp_steps=5, grasp_settle_steps=0, pick_object_part='center', lift_height=0.1, pre_grasp_distance=0.15, approach_direction=tensor([0., 0., -1.]), approach_alignment_max_angle=None, downstream_object_target_poses=(), obj_upright_direction=None, rotate_upright=None, grasp_frame_to_eef=tensor([[1., 0., 0., 0.], [0., 1., 0., 0.], [0., 0., 1., 0.], [0., 0., 0., 1.]]), fixed_object_to_eef=None, target_axis=tensor([0., 0., 1.]))

target_axis: Tensor

Desired world-frame axis, shape (3,) or (B, 3).

embodichain.lab.sim.atomic_actions.primitives.BUILTIN_ACTION_TYPES: tuple[type[AtomicAction], ...] = (<class 'embodichain.lab.sim.atomic_actions.primitives.move_end_effector.MoveEndEffector'>, <class 'embodichain.lab.sim.atomic_actions.primitives.move_joints.MoveJoints'>, <class 'embodichain.lab.sim.atomic_actions.primitives.pick_up.PickUp'>, <class 'embodichain.lab.sim.atomic_actions.primitives.axis_align.AxisAlign'>, <class 'embodichain.lab.sim.atomic_actions.primitives.move_held_object.MoveHeldObject'>, <class 'embodichain.lab.sim.atomic_actions.primitives.place.Place'>, <class 'embodichain.lab.sim.atomic_actions.primitives.pour.Pour'>, <class 'embodichain.lab.sim.atomic_actions.primitives.push_object.PushObject'>, <class 'embodichain.lab.sim.atomic_actions.primitives.press.Press'>, <class 'embodichain.lab.sim.atomic_actions.primitives.slide.Slide'>, <class 'embodichain.lab.sim.atomic_actions.primitives.open_door.OpenDoor'>, <class 'embodichain.lab.sim.atomic_actions.primitives.twist.Twist'>, <class 'embodichain.lab.sim.atomic_actions.primitives.coordinated_pickment.CoordinatedPickment'>, <class 'embodichain.lab.sim.atomic_actions.primitives.coordinated_placement.CoordinatedPlacement'>, <class 'embodichain.lab.sim.atomic_actions.primitives.hand_over.HandOver'>)

Built-in action implementations instantiated once per action engine.

class embodichain.lab.sim.atomic_actions.primitives.CoordinatedPickGoal

Object-centric target for picking and moving one object with two hands.

The left/right grasp poses are not supplied by the caller; they are sampled by the parallel-jaw grasp-pose service at planning time using the dual-arm direction and approach direction declared on CoordinatedPickmentOptions.

Methods:

__init__(semantics, object_target_pose[, ...])

Attributes:

object_initial_pose	Optional initial object pose.
object_target_pose	Target pose for the shared object, shape (4, 4) or (num_envs, 4, 4).

__init__(semantics, object_target_pose, object_initial_pose=None)

object_initial_pose: Tensor | SceneEntityPose | None

Optional initial object pose.

When omitted, the pose is grounded through the semantic object’s stable scene identity.

object_target_pose: Tensor | SceneEntityPose

Target pose for the shared object, shape (4, 4) or (num_envs, 4, 4).

class embodichain.lab.sim.atomic_actions.primitives.CoordinatedPickment

Pick and move a single object pinched by two hands.

Classes:

GoalType	alias of CoordinatedPickGoal
OptionsType	alias of CoordinatedPickmentOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of CoordinatedPickGoal

OptionsType

alias of CoordinatedPickmentOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='left', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.inverse'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)), SkillResourceSlot(slot_id='right', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.inverse'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),))), constraints=(DisjointResourceSlots(slots=('left', 'right')),))

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'coordinated_pickment'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.CoordinatedPickmentOptions

Per-invocation coordinated pickup behavior.

Left/right grasps are sampled from target-local affordance geometry by the engine’s parallel-jaw grasp-pose service. The dual-arm direction splits the object into left/right grasp regions and the approach direction filters the sampled antipodal pairs.

Methods:

__init__([object_motion_keyframes, ...])

Attributes:

approach_direction	World-frame direction used to sample and approach both grasps, shape (3,).
hand_interp_steps	Number of waypoints used for the simultaneous hand-close segment.
hold_steps	Number of waypoints to hold the final object target pose.
left_to_right_arm_direction	World-frame direction from the left arm base to the right arm base, shape (3,).
lift_height	World-Z lift distance before moving to the object target pose.
middle_empty_ratio	Fraction of the object's left-to-right extent left grasp-free in the middle so the two grippers pinch opposite ends.
object_motion_keyframes	Number of object-pose keyframes solved by IK before joint-space interpolation.
pre_grasp_distance	World distance to retreat from each grasp pose along negative TCP z.

__init__(object_motion_keyframes=6, pre_grasp_distance=0.1, lift_height=0.08, hand_interp_steps=10, hold_steps=4, approach_direction=tensor([0., 0., -1.]), left_to_right_arm_direction=tensor([1., 0., 0.]), middle_empty_ratio=0.4)

approach_direction: Tensor

World-frame direction used to sample and approach both grasps, shape (3,).

hand_interp_steps: int

Number of waypoints used for the simultaneous hand-close segment.

hold_steps: int

Number of waypoints to hold the final object target pose.

left_to_right_arm_direction: Tensor

World-frame direction from the left arm base to the right arm base, shape (3,). It partitions the object into left/right grasp regions and should be a finite, non-zero vector; it is normalized at planning time.

lift_height: float

World-Z lift distance before moving to the object target pose.

middle_empty_ratio: float

Fraction of the object’s left-to-right extent left grasp-free in the middle so the two grippers pinch opposite ends. Must be in [0, 1].

object_motion_keyframes: int

Number of object-pose keyframes solved by IK before joint-space interpolation.

pre_grasp_distance: float

World distance to retreat from each grasp pose along negative TCP z.

class embodichain.lab.sim.atomic_actions.primitives.CoordinatedPlacement

Coordinate two held objects: support object below, placing object above.

Classes:

GoalType	alias of CoordinatedPlacementGoal
OptionsType	alias of CoordinatedPlacementOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of CoordinatedPlacementGoal

OptionsType

alias of CoordinatedPlacementOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='placing', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)), SkillResourceSlot(slot_id='support', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),))), constraints=(DisjointResourceSlots(slots=('placing', 'support')),))

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'coordinated_placement'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.CoordinatedPlacementGoal

Object-centric target for dual-arm coordinated placement.

Methods:

__init__(placing_object_target_pose, ...[, ...])

Attributes:

placing_height_offset	World-Z offset above the placing object target pose.
placing_object_target_pose	Target pose for the object released by the placing arm.
release	Whether the placing hand releases.
support_height_offset	World-Z offset above the support object target pose.
support_object_target_pose	Target pose for the object held by the support arm.

__init__(placing_object_target_pose, support_object_target_pose, placing_height_offset=None, support_height_offset=None, release=None)

placing_height_offset: float | None

World-Z offset above the placing object target pose.

placing_object_target_pose: Tensor | SceneEntityPose

Target pose for the object released by the placing arm.

release: bool | None

Whether the placing hand releases. None uses invocation options.

support_height_offset: float | None

World-Z offset above the support object target pose.

support_object_target_pose: Tensor | SceneEntityPose

Target pose for the object held by the support arm.

class embodichain.lab.sim.atomic_actions.primitives.CoordinatedPlacementOptions

Per-invocation coordinated placement behavior.

Methods:

__init__([release, placing_height_offset, ...])

Attributes:

hand_interp_steps	Number of waypoints for the placing-hand release interpolation.
hold_steps	Number of waypoints to hold alignment before releasing.
lift_height	World-Z lift distance for the placing arm after release.
placing_height_offset	Default World-Z offset above the placing object target pose.
release	Whether to open the placing hand at the aligned placement pose.
retreat_steps	Number of waypoints used for the placing-arm lift retreat.
support_height_offset	Default World-Z offset above the support object target pose.

__init__(release=True, placing_height_offset=0.0, support_height_offset=0.0, lift_height=0.08, hand_interp_steps=10, hold_steps=4, retreat_steps=16)

hand_interp_steps: int

Number of waypoints for the placing-hand release interpolation.

hold_steps: int

Number of waypoints to hold alignment before releasing.

lift_height: float

World-Z lift distance for the placing arm after release.

placing_height_offset: float

Default World-Z offset above the placing object target pose.

release: bool

Whether to open the placing hand at the aligned placement pose.

retreat_steps: int

Number of waypoints used for the placing-arm lift retreat.

support_height_offset: float

Default World-Z offset above the support object target pose.

class embodichain.lab.sim.atomic_actions.primitives.EndEffectorPoseGoal

End-effector pose goal with optional batched intermediate waypoints.

Methods:

__init__(xpos)

Attributes:

xpos	Homogeneous pose with shape (4,4), (B,4,4) or (B,N,4,4).

__init__(xpos)

xpos: Tensor | SceneEntityPose

Homogeneous pose with shape (4,4), (B,4,4) or (B,N,4,4).

class embodichain.lab.sim.atomic_actions.primitives.GraspGoal

Pickup target with an affordance-selected or supplied grasp pose.

Methods:

__init__(semantics[, grasp_xpos])

Attributes:

grasp_xpos

Optional end-effector grasp pose.

__init__(semantics, grasp_xpos=None)

grasp_xpos: Tensor | SceneEntityPose | None

Optional end-effector grasp pose.

When omitted, PickUp uses the configured fixed object-relative grasp when available, otherwise it selects one from the target affordance. An explicit tensor or late-bound SceneEntityPose skips grasp sampling. Late-bound poses also declare the scene dependency used by closed-loop execution recovery.

class embodichain.lab.sim.atomic_actions.primitives.HandOver

Pick an object with the nearer arm, hand it over, and place it.

For each environment, the action chooses the arm whose root link is closer to the observed object pose. It samples at most 1000 mesh-surface points and applies SVD in the current object pose to find obj_longest_axis. When that axis is closer to world Z than to the horizontal plane, both grasp approaches point toward the object horizontally and tilt downward by 45 degrees. Otherwise both approaches are world-Z downward.

The first arm grasps the projected end of obj_longest_axis nearest its current TCP; the receiving arm grasps the opposite end at the predicted middle object pose. This keeps the two hands from selecting the same object region regardless of whether a long object is standing or lying down.

After each grasp waypoint, subsequent EEF waypoints preserve that grasp rotation and change translation only. In particular, placement first moves strictly horizontally at the handover height and then lowers to the final target pose before releasing the object.

Classes:

GoalType	alias of HandOverGoal
OptionsType	alias of HandOverOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of HandOverGoal

OptionsType

alias of HandOverOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='source', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)), SkillResourceSlot(slot_id='destination', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),))), constraints=(DisjointResourceSlots(slots=('source', 'destination')),))

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'hand_over'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.HandOverGoal

Object to pick and hand over, plus its final object pose.

Methods:

__init__(semantics, target_pose)

Attributes:

target_pose

Final object pose after the receiving arm lowers and releases it.

__init__(semantics, target_pose)

target_pose: PoseGoalValue

Final object pose after the receiving arm lowers and releases it.

class embodichain.lab.sim.atomic_actions.primitives.HandOverOptions

Per-invocation pick-up, handover, and placement behavior.

Methods:

__init__([pre_grasp_distance, lift_height, ...])

Attributes:

hand_interp_steps	Waypoints used by every gripper open/close interpolation.
lift_height	World-Z distance used to lift the object after the first grasp.
pre_grasp_distance	Distance from each grasp pose to its approach pose, in metres.

__init__(pre_grasp_distance=0.1, lift_height=0.1, hand_interp_steps=10)

hand_interp_steps: int

Waypoints used by every gripper open/close interpolation.

lift_height: float

World-Z distance used to lift the object after the first grasp.

pre_grasp_distance: float

Distance from each grasp pose to its approach pose, in metres.

class embodichain.lab.sim.atomic_actions.primitives.HeldObjectPoseGoal

Desired pose for the object held by this action’s control part.

Methods:

__init__(object_target_pose)

Attributes:

object_target_pose

Target object pose, shape (4, 4) or (num_envs, 4, 4).

__init__(object_target_pose)

object_target_pose: Tensor | SceneEntityPose

Target object pose, shape (4, 4) or (num_envs, 4, 4).

class embodichain.lab.sim.atomic_actions.primitives.JointPositionGoal

Explicit or named joint-space goal for a bound robot resource.

Methods:

__init__(target)

Attributes:

target

Joint qpos/waypoints or a named control-part profile command.

__init__(target)

target: Tensor | str

Joint qpos/waypoints or a named control-part profile command.

class embodichain.lab.sim.atomic_actions.primitives.MoveEndEffector

Plan a free-space move for a bound manipulator.

Classes:

GoalType	alias of EndEffectorPoseGoal
OptionsType	alias of MoveEndEffectorOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of EndEffectorPoseGoal

OptionsType

alias of MoveEndEffectorOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})),), constraints=()),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'move_end_effector'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.MoveEndEffectorOptions

Per-invocation behavior for MoveEndEffector.

Methods:

__init__()

__init__()

class embodichain.lab.sim.atomic_actions.primitives.MoveHeldObject

Move the held object to the exact target object pose with a closed hand.

The requested object orientation is preserved exactly. Callers that need a transport orientation must encode it in HeldObjectPoseGoal; this action never substitutes an implicit end-effector orientation.

Classes:

GoalType	alias of HeldObjectPoseGoal
OptionsType	alias of MoveHeldObjectOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of HeldObjectPoseGoal

OptionsType

alias of MoveHeldObjectOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'move_held_object'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.MoveHeldObjectOptions

Per-invocation held-object transport behavior.

Methods:

__init__()

__init__()

class embodichain.lab.sim.atomic_actions.primitives.MoveJoints

Plan joint motion from the observed state to one or more waypoints.

Classes:

GoalType	alias of JointPositionGoal
OptionsType	alias of MoveJointsOptions

Attributes:

agent_visible	Whether an Action Agent should expose this skill by default.
binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of JointPositionGoal

OptionsType

alias of MoveJointsOptions

agent_visible: ClassVar[bool] = False

Whether an Action Agent should expose this skill by default.

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.joint_position'}), required_commands=mappingproxy({})),), constraints=()),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'move_joints'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.MoveJointsOptions

Per-invocation behavior for MoveJoints.

Methods:

__init__()

__init__()

class embodichain.lab.sim.atomic_actions.primitives.OpenDoor

Approach, grasp, rotate a door about its hinge, release, and retract.

Classes:

GoalType	alias of OpenDoorGoal
OptionsType	alias of OpenDoorOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of OpenDoorGoal

OptionsType

alias of OpenDoorOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'open_door'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.OpenDoorGoal

Door handle and desired absolute opening state.

Methods:

__init__(semantics, target_pose, open_fraction)

Attributes:

open_fraction	Desired hinge position normalized from its closed to open legal endpoint.
target_pose	Handle-link pose snapshot or late-bound scene-entity reference.

__init__(semantics, target_pose, open_fraction)

open_fraction: float | Tensor

Desired hinge position normalized from its closed to open legal endpoint.

target_pose: Tensor | SceneEntityPose

Handle-link pose snapshot or late-bound scene-entity reference.

class embodichain.lab.sim.atomic_actions.primitives.OpenDoorOptions

Per-invocation approach, interpolation, release, and retract behavior.

Methods:

__init__([hand_interp_steps, ...])

Attributes:

approach_distance	Pre-grasp distance opposite the automatically inferred approach axis.
door_waypoint_count	Number of Cartesian keyframes along the handle's circular arc.
hand_interp_steps	Number of waypoints used for each close/open hand segment.
joint_position_tolerance	Tolerance for legal-limit and already-open comparisons in radians.
retract_distance	Post-release retreat distance opposite the rotated approach axis.

__init__(hand_interp_steps=5, door_waypoint_count=20, approach_distance=0.1, retract_distance=0.1, joint_position_tolerance=0.0001)

approach_distance: float

Pre-grasp distance opposite the automatically inferred approach axis.

door_waypoint_count: int

Number of Cartesian keyframes along the handle’s circular arc.

hand_interp_steps: int

Number of waypoints used for each close/open hand segment.

joint_position_tolerance: float

Tolerance for legal-limit and already-open comparisons in radians.

retract_distance: float

Post-release retreat distance opposite the rotated approach axis.

class embodichain.lab.sim.atomic_actions.primitives.PickUp

Approach a grasp pose, close the gripper, lift.

Classes:

GoalType	alias of GraspGoal
OptionsType	alias of PickUpOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of GraspGoal

OptionsType

alias of PickUpOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'kinematics.batch_inverse', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'pick_up'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.PickUpOptions

Per-invocation pickup behavior.

Methods:

__init__([hand_interp_steps, ...])

Attributes:

approach_alignment_max_angle	Optional maximum TCP z-axis deviation from the approach direction.
approach_direction	World-frame direction from the pre-grasp pose to the grasp pose.
downstream_object_target_poses	Future object poses that must be reachable with the selected grasp.
fixed_object_to_eef	Optional object-frame to end-effector SE(3) grasp calibration.
grasp_frame_to_eef	Canonical grasp-frame to robot end-effector SE(3) calibration.
grasp_settle_steps	Fully closed hold frames before lifting the end-effector.
hand_interp_steps	Number of waypoints for the gripper-close interpolation segment.
lift_height	Height (m) to lift the end-effector after closing the gripper.
obj_upright_direction	Optional object local direction used to choose the upright grasp rotation.
pick_object_part	Name of the object part to pick up (used for grasp pose generation).
pre_grasp_distance	Distance to offset back from the grasp pose along the approach direction.
rotate_upright	Optional rotation (radians) about the grasp x-axis to apply after grasp selection.

__init__(hand_interp_steps=5, grasp_settle_steps=0, pick_object_part='center', lift_height=0.1, pre_grasp_distance=0.15, approach_direction=tensor([0., 0., -1.]), approach_alignment_max_angle=None, downstream_object_target_poses=(), obj_upright_direction=None, rotate_upright=None, grasp_frame_to_eef=tensor([[1., 0., 0., 0.], [0., 1., 0., 0.], [0., 0., 1., 0.], [0., 0., 0., 1.]]), fixed_object_to_eef=None)

approach_alignment_max_angle: float | None

Optional maximum TCP z-axis deviation from the approach direction.

approach_direction: Tensor

World-frame direction from the pre-grasp pose to the grasp pose.

downstream_object_target_poses: tuple[Tensor | SceneEntityPose, ...]

Future object poses that must be reachable with the selected grasp.

fixed_object_to_eef: Tensor | None

Optional object-frame to end-effector SE(3) grasp calibration.

When no explicit goal grasp is supplied, this transform bypasses affordance sampling and the sampled-grasp orientation/calibration adjustments.

grasp_frame_to_eef: Tensor

Canonical grasp-frame to robot end-effector SE(3) calibration.

grasp_settle_steps: int

Fully closed hold frames before lifting the end-effector.

hand_interp_steps: int

Number of waypoints for the gripper-close interpolation segment.

lift_height: float

Height (m) to lift the end-effector after closing the gripper.

obj_upright_direction: Tensor | None

Optional object local direction used to choose the upright grasp rotation.

pick_object_part: str

Name of the object part to pick up (used for grasp pose generation). Currently support [center | top | bottom].

pre_grasp_distance: float

Distance to offset back from the grasp pose along the approach direction.

rotate_upright: float | None

Optional rotation (radians) about the grasp x-axis to apply after grasp selection.

class embodichain.lab.sim.atomic_actions.primitives.Place

Lower the held object to a place pose, open the gripper, retract.

The PlaceGoal may carry either a single waypoint (num_envs, 4, 4) (or a broadcastable (4, 4)) or a multi-waypoint trajectory (num_envs, n_waypoint, 4, 4). In the multi-waypoint case the approach segment visits every waypoint in order; approaching from above the first waypoint, descending through each waypoint, then opening the gripper at the final waypoint and retracting to above the last waypoint. Starting joint positions are inherited from PlanningContext.

An AssembleGoal replaces the explicit EEF pose with an assembly affordance: the place pose is derived from the base object’s snapshot pose and assemble_to_base_pose, converted to an EEF pose through the held object’s object_to_eef (read from PlanningContext).

Attributes:

GoalType	Concrete goal dataclass or dataclasses accepted by this skill.
binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

Classes:

OptionsType

alias of PlaceOptions

GoalType: ClassVar[type | tuple[type, ...]] = (<class 'embodichain.lab.sim.atomic_actions.primitives.place.PlaceGoal'>, <class 'embodichain.lab.sim.atomic_actions.primitives.place.AssembleGoal'>)

Concrete goal dataclass or dataclasses accepted by this skill.

OptionsType

alias of PlaceOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'place'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.PlaceGoal

End-effector release-pose target used by Place.

Methods:

__init__(xpos[, tcp_symmetry])

Attributes:

tcp_symmetry	Optional TCP-frame symmetry allowed by the placement semantics.
xpos	Target end-effector release pose.

__init__(xpos, tcp_symmetry='none')

tcp_symmetry: Literal['none', 'z_roll_180']

Optional TCP-frame symmetry allowed by the placement semantics.

"none" preserves the pose exactly. "z_roll_180" lets placement choose between the pose and its TCP z-roll 180 equivalent, which flips TCP x/y while preserving TCP z and translation.

xpos: Tensor | SceneEntityPose

Target end-effector release pose.

Accepts (4, 4), (num_envs, 4, 4), or (num_envs, n_waypoint, 4, 4).

class embodichain.lab.sim.atomic_actions.primitives.PlaceOptions

Per-invocation placement behavior.

Methods:

__init__([hand_interp_steps, ...])

Attributes:

cartesian_waypoint_count	Number of fixed-orientation Cartesian keyframes per translation segment.
hand_interp_steps	Number of waypoints for the gripper-open interpolation segment.
lift_height	Height (m) to retract the end-effector after opening the gripper.
max_approach_retract_z	Optional maximum world-frame TCP z for approach and retract poses (m).
preserve_current_object_orientation	Keep the held object's observed world orientation at the place target.
release_settle_steps	Fully open hold frames before retracting the end-effector.

__init__(hand_interp_steps=5, release_settle_steps=0, lift_height=0.1, max_approach_retract_z=None, cartesian_waypoint_count=1, preserve_current_object_orientation=False)

cartesian_waypoint_count: int

Number of fixed-orientation Cartesian keyframes per translation segment.

hand_interp_steps: int

Number of waypoints for the gripper-open interpolation segment.

lift_height: float

Height (m) to retract the end-effector after opening the gripper.

max_approach_retract_z: float | None

Optional maximum world-frame TCP z for approach and retract poses (m).

preserve_current_object_orientation: bool

Keep the held object’s observed world orientation at the place target.

release_settle_steps: int

Fully open hold frames before retracting the end-effector.

class embodichain.lab.sim.atomic_actions.primitives.Pour

Rotate and return an exclusively held object about its internal axis.

Classes:

GoalType	alias of PourGoal
OptionsType	alias of PourOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of PourGoal

OptionsType

alias of PourOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'pour'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.PourGoal

Rotate the object currently held by the bound manipulator.

Methods:

__init__()

__init__()

class embodichain.lab.sim.atomic_actions.primitives.PourOptions

Per-invocation pouring behavior.

Methods:

__init__([rotate_angle])

Attributes:

rotate_angle

Signed rotation about the held object's local internal axis, in radians.

__init__(rotate_angle=0.7853981633974483)

rotate_angle: float

Signed rotation about the held object’s local internal axis, in radians.

class embodichain.lab.sim.atomic_actions.primitives.Press

Open-loop motion primitive that approaches, presses, and retracts.

Classes:

GoalType	alias of PressGoal
OptionsType	alias of PressOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of PressGoal

OptionsType

alias of PressOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'press'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.PressGoal

Target object described by a press affordance.

Methods:

__init__(semantics, target_pose)

Attributes:

target_pose

Target pose snapshot or late-bound stable scene-entity reference.

__init__(semantics, target_pose)

target_pose: Tensor | SceneEntityPose

Target pose snapshot or late-bound stable scene-entity reference.

class embodichain.lab.sim.atomic_actions.primitives.PressOptions

Per-invocation pressing behavior.

Methods:

__init__([hand_interp_steps, ...])

Attributes:

approach_distance	Distance from the press position opposite the press direction.
hand_interp_steps	Number of waypoints used to close the hand.
press_distance	Distance traveled into the target along its press axis.
press_position	Optional local-frame position overriding the affordance press position.

__init__(hand_interp_steps=5, approach_distance=0.1, press_distance=0.05, press_position=None)

approach_distance: float

Distance from the press position opposite the press direction.

hand_interp_steps: int

Number of waypoints used to close the hand.

press_distance: float

Distance traveled into the target along its press axis.

press_position: tuple[float, float, float] | None

Optional local-frame position overriding the affordance press position.

class embodichain.lab.sim.atomic_actions.primitives.PushObject

Close the end effector, contact a rigid object, and push it in-plane.

Classes:

GoalType	alias of PushObjectGoal
OptionsType	alias of PushObjectOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of PushObjectGoal

OptionsType

alias of PushObjectOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'push_object'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.PushObjectGoal

Push one rigid object toward a target pose on the target support plane.

Methods:

__init__(semantics, target_pose)

Attributes:

target_pose

Desired object pose or a late-bound scene-entity target reference.

__init__(semantics, target_pose)

target_pose: PoseGoalValue

Desired object pose or a late-bound scene-entity target reference.

class embodichain.lab.sim.atomic_actions.primitives.PushObjectOptions

Per-invocation planar pushing behavior.

Methods:

__init__([hand_interp_steps, ...])

Attributes:

approach_height	Distance above the contact pose used for the free-space approach.
completion_tolerance	Planar target distance at which the action succeeds without moving.
contact_distance	Initial planar clearance behind the object along the push direction.
contact_frame_to_eef	Contact-frame to robot end-effector SE(3) calibration.
hand_interp_steps	Number of waypoints used to close the end effector before approach.
object_contact_offset	Object-local point used as the center of the planar contact frame.
push_overshoot	Additional end-effector travel beyond the object's target displacement.
retract_height	Distance above the pushed pose used for the final retraction.
support_frame_planar_contact_offset	Optional target-support-frame override for the contact's planar offset.
tool_calibrations	Per-control-part tool-frame overrides for asymmetric robot arms.

__init__(hand_interp_steps=5, approach_height=0.1, retract_height=0.1, contact_distance=0.03, push_overshoot=0.0, completion_tolerance=0.0, object_contact_offset=tensor([0., 0., 0.]), support_frame_planar_contact_offset=None, contact_frame_to_eef=tensor([[1., 0., 0., 0.], [0., 1., 0., 0.], [0., 0., 1., 0.], [0., 0., 0., 1.]]), tool_calibrations=())

approach_height: float

Distance above the contact pose used for the free-space approach.

completion_tolerance: float

Planar target distance at which the action succeeds without moving.

contact_distance: float

Initial planar clearance behind the object along the push direction.

contact_frame_to_eef: Tensor

Contact-frame to robot end-effector SE(3) calibration.

hand_interp_steps: int

Number of waypoints used to close the end effector before approach.

object_contact_offset: Tensor

Object-local point used as the center of the planar contact frame.

push_overshoot: float

Additional end-effector travel beyond the object’s target displacement.

retract_height: float

Distance above the pushed pose used for the final retraction.

support_frame_planar_contact_offset: Tensor | None

Optional target-support-frame override for the contact’s planar offset.

tool_calibrations: tuple[PushObjectToolCalibration, ...]

Per-control-part tool-frame overrides for asymmetric robot arms.

class embodichain.lab.sim.atomic_actions.primitives.PushObjectToolCalibration

End-effector calibration selected by a bound motion control part.

Parameters:

• control_part (str) – Exact motion control-part identifier that selects this calibration.

• contact_frame_to_eef (Tensor) – Contact-frame to robot end-effector SE(3) calibration.

• contact_distance (float | None) – Optional tool-specific planar clearance behind the object. None uses PushObjectOptions’ default.

Methods:

__init__(control_part, contact_frame_to_eef)

Attributes:

contact_distance	Optional tool-specific planar clearance behind the object.
contact_frame_to_eef	Contact-frame to robot end-effector SE(3) calibration.
control_part	Exact motion control-part identifier that selects this calibration.

__init__(control_part, contact_frame_to_eef, contact_distance=None)

contact_distance: float | None

Optional tool-specific planar clearance behind the object.

contact_frame_to_eef: Tensor

Contact-frame to robot end-effector SE(3) calibration.

control_part: str

Exact motion control-part identifier that selects this calibration.

class embodichain.lab.sim.atomic_actions.primitives.Slide

Open-loop approach, grasp, and axis-constrained sliding motion.

Classes:

GoalType	alias of SlideGoal
OptionsType	alias of SlideOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of SlideGoal

OptionsType

alias of SlideOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'slide'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.SlideGoal

Translating articulation link described by a slide affordance.

Methods:

__init__(semantics, target_pose)

Attributes:

target_pose

Link pose snapshot or late-bound stable scene-entity reference.

__init__(semantics, target_pose)

target_pose: Tensor | SceneEntityPose

Link pose snapshot or late-bound stable scene-entity reference.

class embodichain.lab.sim.atomic_actions.primitives.SlideOptions

Per-invocation sliding behavior for a translating articulation link.

Methods:

__init__([direction, hand_interp_steps, ...])

Attributes:

approach_distance	Pre-grasp distance opposite the approach/push axis.
direction	Whether to pull the part open or push it closed.
hand_interp_steps	Number of waypoints used for each close/open hand segment.
translation_distance	Distance traveled along the pull or push direction.

__init__(direction='pull', hand_interp_steps=5, approach_distance=0.1, translation_distance=0.15)

approach_distance: float

Pre-grasp distance opposite the approach/push axis.

direction: Literal['pull', 'push']

Whether to pull the part open or push it closed.

hand_interp_steps: int

Number of waypoints used for each close/open hand segment.

translation_distance: float

Distance traveled along the pull or push direction.

class embodichain.lab.sim.atomic_actions.primitives.Twist

Open-loop approach, grasp, twist, release, and retract motion.

Classes:

GoalType	alias of TwistGoal
OptionsType	alias of TwistOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of TwistGoal

OptionsType

alias of TwistOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'twist'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.TwistGoal

Target object described by a twist affordance.

Methods:

__init__(semantics, target_pose)

Attributes:

target_pose

Target pose snapshot or late-bound stable scene-entity reference.

__init__(semantics, target_pose)

target_pose: Tensor | SceneEntityPose

Target pose snapshot or late-bound stable scene-entity reference.

class embodichain.lab.sim.atomic_actions.primitives.TwistOptions

Per-invocation twisting behavior.

Methods:

__init__([hand_interp_steps, ...])

Attributes:

hand_interp_steps	Number of waypoints used for each close/open hand segment.
pre_grasp_distance	Distance from the grasp pose along its negative z-axis.
twist_angle	Requested twist rotation in radians.
twist_waypoint_count	Number of Cartesian keyframes along the target's circular twist arc.

__init__(hand_interp_steps=5, twist_waypoint_count=8, pre_grasp_distance=0.1, twist_angle=0.7853981633974483)

hand_interp_steps: int

Number of waypoints used for each close/open hand segment.

pre_grasp_distance: float

Distance from the grasp pose along its negative z-axis.

twist_angle: float

Requested twist rotation in radians.

twist_waypoint_count: int

Number of Cartesian keyframes along the target’s circular twist arc.

Overview

Concrete implementations of the built-in atomic-action primitives. Each primitive is an AtomicAction that accepts a ResolvedActionRequest and a PlanningContext. Planning returns a side-effect-free ActionPlan with a full-robot timed trajectory and uncommitted expected effects.

Built-in Primitive Actions

MoveEndEffector	Plan a free-space move for a bound manipulator.
MoveEndEffectorOptions	Per-invocation behavior for MoveEndEffector.
MoveJoints	Plan joint motion from the observed state to one or more waypoints.
MoveJointsOptions	Per-invocation behavior for MoveJoints.
PickUp	Approach a grasp pose, close the gripper, lift.
PickUpOptions	Per-invocation pickup behavior.
AxisAlign	Grasp an object and align its local axis to a world axis.
AxisAlignOptions	Per-invocation grasp-and-axis-alignment behavior.
MoveHeldObject	Move the held object to the exact target object pose with a closed hand.
MoveHeldObjectOptions	Per-invocation held-object transport behavior.
Pour	Rotate and return an exclusively held object about its internal axis.
PourOptions	Per-invocation pouring behavior.
PushObject	Close the end effector, contact a rigid object, and push it in-plane.
PushObjectOptions	Per-invocation planar pushing behavior.
PushObjectToolCalibration	End-effector calibration selected by a bound motion control part.
Place	Lower the held object to a place pose, open the gripper, retract.
PlaceOptions	Per-invocation placement behavior.
Press	Open-loop motion primitive that approaches, presses, and retracts.
PressOptions	Per-invocation pressing behavior.
Slide	Open-loop approach, grasp, and axis-constrained sliding motion.
SlideOptions	Per-invocation sliding behavior for a translating articulation link.
OpenDoor	Approach, grasp, rotate a door about its hinge, release, and retract.
OpenDoorOptions	Per-invocation approach, interpolation, release, and retract behavior.
Twist	Open-loop approach, grasp, twist, release, and retract motion.
TwistOptions	Per-invocation twisting behavior.
CoordinatedPickment	Pick and move a single object pinched by two hands.
CoordinatedPickmentOptions	Per-invocation coordinated pickup behavior.
CoordinatedPlacement	Coordinate two held objects: support object below, placing object above.
CoordinatedPlacementOptions	Per-invocation coordinated placement behavior.
HandOver	Pick an object with the nearer arm, hand it over, and place it.
HandOverGoal	Object to pick and hand over, plus its final object pose.
HandOverOptions	Per-invocation pick-up, handover, and placement behavior.

Built-in Goal Contracts

EndEffectorPoseGoal	End-effector pose goal with optional batched intermediate waypoints.
JointPositionGoal	Explicit or named joint-space goal for a bound robot resource.
GraspGoal	Pickup target with an affordance-selected or supplied grasp pose.
AxisAlignGoal	Object whose local axis should be aligned after an antipodal grasp.
HeldObjectPoseGoal	Desired pose for the object held by this action's control part.
PourGoal	Rotate the object currently held by the bound manipulator.
PushObjectGoal	Push one rigid object toward a target pose on the target support plane.
PlaceGoal	End-effector release-pose target used by Place.
AssembleGoal	Place a held assemble object onto a base object at a relative pose.
PressGoal	Target object described by a press affordance.
SlideGoal	Translating articulation link described by a slide affordance.
OpenDoorGoal	Door handle and desired absolute opening state.
TwistGoal	Target object described by a twist affordance.
CoordinatedPickGoal	Object-centric target for picking and moving one object with two hands.
CoordinatedPlacementGoal	Object-centric target for dual-arm coordinated placement.
HandOverGoal	Object to pick and hand over, plus its final object pose.

MoveEndEffector

MoveEndEffector atomic action implementation.

Classes:

EndEffectorPoseGoal	End-effector pose goal with optional batched intermediate waypoints.
MoveEndEffector	Plan a free-space move for a bound manipulator.
MoveEndEffectorOptions	Per-invocation behavior for MoveEndEffector.

class embodichain.lab.sim.atomic_actions.primitives.move_end_effector.EndEffectorPoseGoal

Bases: object

End-effector pose goal with optional batched intermediate waypoints.

Attributes:

xpos	Homogeneous pose with shape (4,4), (B,4,4) or (B,N,4,4).

xpos: Tensor | SceneEntityPose

Homogeneous pose with shape (4,4), (B,4,4) or (B,N,4,4).

class embodichain.lab.sim.atomic_actions.primitives.move_end_effector.MoveEndEffector

Bases: AtomicAction[EndEffectorPoseGoal, MoveEndEffectorOptions]

Plan a free-space move for a bound manipulator.

Classes:

GoalType	alias of EndEffectorPoseGoal
OptionsType	alias of MoveEndEffectorOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of EndEffectorPoseGoal

OptionsType

alias of MoveEndEffectorOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})),), constraints=()),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'move_end_effector'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.move_end_effector.MoveEndEffectorOptions

Bases: ActionOptions

Per-invocation behavior for MoveEndEffector.

MoveJoints

MoveJoints atomic action implementation.

Classes:

JointPositionGoal	Explicit or named joint-space goal for a bound robot resource.
MoveJoints	Plan joint motion from the observed state to one or more waypoints.
MoveJointsOptions	Per-invocation behavior for MoveJoints.

class embodichain.lab.sim.atomic_actions.primitives.move_joints.JointPositionGoal

Bases: object

Explicit or named joint-space goal for a bound robot resource.

Attributes:

target

Joint qpos/waypoints or a named control-part profile command.

target: Tensor | str

Joint qpos/waypoints or a named control-part profile command.

class embodichain.lab.sim.atomic_actions.primitives.move_joints.MoveJoints

Bases: AtomicAction[JointPositionGoal, MoveJointsOptions]

Plan joint motion from the observed state to one or more waypoints.

Classes:

GoalType	alias of JointPositionGoal
OptionsType	alias of MoveJointsOptions

Attributes:

agent_visible	Whether an Action Agent should expose this skill by default.
binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of JointPositionGoal

OptionsType

alias of MoveJointsOptions

agent_visible: ClassVar[bool] = False

Whether an Action Agent should expose this skill by default.

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.joint_position'}), required_commands=mappingproxy({})),), constraints=()),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'move_joints'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.move_joints.MoveJointsOptions

Bases: ActionOptions

Per-invocation behavior for MoveJoints.

PickUp

PickUp atomic action implementation.

Classes:

GraspGoal	Pickup target with an affordance-selected or supplied grasp pose.
PickUp	Approach a grasp pose, close the gripper, lift.
PickUpOptions	Per-invocation pickup behavior.

class embodichain.lab.sim.atomic_actions.primitives.pick_up.GraspGoal

Bases: ObjectActionGoal

Pickup target with an affordance-selected or supplied grasp pose.

Attributes:

grasp_xpos

Optional end-effector grasp pose.

grasp_xpos: Tensor | SceneEntityPose | None

Optional end-effector grasp pose.

When omitted, PickUp uses the configured fixed object-relative grasp when available, otherwise it selects one from the target affordance. An explicit tensor or late-bound SceneEntityPose skips grasp sampling. Late-bound poses also declare the scene dependency used by closed-loop execution recovery.

class embodichain.lab.sim.atomic_actions.primitives.pick_up.PickUp

Bases: AtomicAction[GraspGoal, PickUpOptions]

Approach a grasp pose, close the gripper, lift.

Classes:

GoalType	alias of GraspGoal
OptionsType	alias of PickUpOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of GraspGoal

OptionsType

alias of PickUpOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'kinematics.batch_inverse', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'pick_up'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.pick_up.PickUpOptions

Bases: ActionOptions

Per-invocation pickup behavior.

Attributes:

approach_alignment_max_angle	Optional maximum TCP z-axis deviation from the approach direction.
approach_direction	World-frame direction from the pre-grasp pose to the grasp pose.
downstream_object_target_poses	Future object poses that must be reachable with the selected grasp.
fixed_object_to_eef	Optional object-frame to end-effector SE(3) grasp calibration.
grasp_frame_to_eef	Canonical grasp-frame to robot end-effector SE(3) calibration.
grasp_settle_steps	Fully closed hold frames before lifting the end-effector.
hand_interp_steps	Number of waypoints for the gripper-close interpolation segment.
lift_height	Height (m) to lift the end-effector after closing the gripper.
obj_upright_direction	Optional object local direction used to choose the upright grasp rotation.
pick_object_part	Name of the object part to pick up (used for grasp pose generation).
pre_grasp_distance	Distance to offset back from the grasp pose along the approach direction.
rotate_upright	Optional rotation (radians) about the grasp x-axis to apply after grasp selection.

approach_alignment_max_angle: float | None

Optional maximum TCP z-axis deviation from the approach direction.

approach_direction: Tensor

World-frame direction from the pre-grasp pose to the grasp pose.

downstream_object_target_poses: tuple[Tensor | SceneEntityPose, ...]

Future object poses that must be reachable with the selected grasp.

fixed_object_to_eef: Tensor | None

Optional object-frame to end-effector SE(3) grasp calibration.

When no explicit goal grasp is supplied, this transform bypasses affordance sampling and the sampled-grasp orientation/calibration adjustments.

grasp_frame_to_eef: Tensor

Canonical grasp-frame to robot end-effector SE(3) calibration.

grasp_settle_steps: int

Fully closed hold frames before lifting the end-effector.

hand_interp_steps: int

Number of waypoints for the gripper-close interpolation segment.

lift_height: float

Height (m) to lift the end-effector after closing the gripper.

obj_upright_direction: Tensor | None

Optional object local direction used to choose the upright grasp rotation.

pick_object_part: str

Name of the object part to pick up (used for grasp pose generation). Currently support [center | top | bottom].

pre_grasp_distance: float

Distance to offset back from the grasp pose along the approach direction.

rotate_upright: float | None

Optional rotation (radians) about the grasp x-axis to apply after grasp selection.

AxisAlign

AxisAlign atomic action implementation.

Classes:

AxisAlign	Grasp an object and align its local axis to a world axis.
AxisAlignGoal	Object whose local axis should be aligned after an antipodal grasp.
AxisAlignOptions	Per-invocation grasp-and-axis-alignment behavior.

class embodichain.lab.sim.atomic_actions.primitives.axis_align.AxisAlign

Bases: AtomicAction[AxisAlignGoal, AxisAlignOptions]

Grasp an object and align its local axis to a world axis.

Classes:

GoalType	alias of AxisAlignGoal
OptionsType	alias of AxisAlignOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of AxisAlignGoal

OptionsType

alias of AxisAlignOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'axis_align'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.axis_align.AxisAlignGoal

Bases: ObjectActionGoal

Object whose local axis should be aligned after an antipodal grasp.

Attributes:

grasp_xpos

Optional explicit end-effector grasp pose; omitted poses are sampled.

grasp_xpos: Tensor | SceneEntityPose | None

Optional explicit end-effector grasp pose; omitted poses are sampled.

class embodichain.lab.sim.atomic_actions.primitives.axis_align.AxisAlignOptions

Bases: PickUpOptions

Per-invocation grasp-and-axis-alignment behavior.

Attributes:

target_axis

Desired world-frame axis, shape (3,) or (B, 3).

target_axis: Tensor

Desired world-frame axis, shape (3,) or (B, 3).

MoveHeldObject

MoveHeldObject atomic action implementation.

Classes:

HeldObjectPoseGoal	Desired pose for the object held by this action's control part.
MoveHeldObject	Move the held object to the exact target object pose with a closed hand.
MoveHeldObjectOptions	Per-invocation held-object transport behavior.

class embodichain.lab.sim.atomic_actions.primitives.move_held_object.HeldObjectPoseGoal

Bases: object

Desired pose for the object held by this action’s control part.

Attributes:

object_target_pose

Target object pose, shape (4, 4) or (num_envs, 4, 4).

object_target_pose: Tensor | SceneEntityPose

Target object pose, shape (4, 4) or (num_envs, 4, 4).

class embodichain.lab.sim.atomic_actions.primitives.move_held_object.MoveHeldObject

Bases: AtomicAction[HeldObjectPoseGoal, MoveHeldObjectOptions]

Move the held object to the exact target object pose with a closed hand.

The requested object orientation is preserved exactly. Callers that need a transport orientation must encode it in HeldObjectPoseGoal; this action never substitutes an implicit end-effector orientation.

Classes:

GoalType	alias of HeldObjectPoseGoal
OptionsType	alias of MoveHeldObjectOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of HeldObjectPoseGoal

OptionsType

alias of MoveHeldObjectOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'move_held_object'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.move_held_object.MoveHeldObjectOptions

Bases: ActionOptions

Per-invocation held-object transport behavior.

Pour

Pour atomic action implementation.

Classes:

Pour	Rotate and return an exclusively held object about its internal axis.
PourGoal	Rotate the object currently held by the bound manipulator.
PourOptions	Per-invocation pouring behavior.

class embodichain.lab.sim.atomic_actions.primitives.pour.Pour

Bases: AtomicAction[PourGoal, PourOptions]

Rotate and return an exclusively held object about its internal axis.

Classes:

GoalType	alias of PourGoal
OptionsType	alias of PourOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of PourGoal

OptionsType

alias of PourOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'pour'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.pour.PourGoal

Bases: object

Rotate the object currently held by the bound manipulator.

class embodichain.lab.sim.atomic_actions.primitives.pour.PourOptions

Bases: ActionOptions

Per-invocation pouring behavior.

Attributes:

rotate_angle

Signed rotation about the held object's local internal axis, in radians.

rotate_angle: float

Signed rotation about the held object’s local internal axis, in radians.

PushObject

Planar rigid-object pushing atomic action.

Classes:

PushObject	Close the end effector, contact a rigid object, and push it in-plane.
PushObjectGoal	Push one rigid object toward a target pose on the target support plane.
PushObjectOptions	Per-invocation planar pushing behavior.
PushObjectToolCalibration	End-effector calibration selected by a bound motion control part.

class embodichain.lab.sim.atomic_actions.primitives.push_object.PushObject

Bases: AtomicAction[PushObjectGoal, PushObjectOptions]

Close the end effector, contact a rigid object, and push it in-plane.

Classes:

GoalType	alias of PushObjectGoal
OptionsType	alias of PushObjectOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of PushObjectGoal

OptionsType

alias of PushObjectOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'push_object'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.push_object.PushObjectGoal

Bases: ObjectActionGoal

Push one rigid object toward a target pose on the target support plane.

Attributes:

target_pose

Desired object pose or a late-bound scene-entity target reference.

target_pose: PoseGoalValue

Desired object pose or a late-bound scene-entity target reference.

class embodichain.lab.sim.atomic_actions.primitives.push_object.PushObjectOptions

Bases: ActionOptions

Per-invocation planar pushing behavior.

Attributes:

approach_height	Distance above the contact pose used for the free-space approach.
completion_tolerance	Planar target distance at which the action succeeds without moving.
contact_distance	Initial planar clearance behind the object along the push direction.
contact_frame_to_eef	Contact-frame to robot end-effector SE(3) calibration.
hand_interp_steps	Number of waypoints used to close the end effector before approach.
object_contact_offset	Object-local point used as the center of the planar contact frame.
push_overshoot	Additional end-effector travel beyond the object's target displacement.
retract_height	Distance above the pushed pose used for the final retraction.
support_frame_planar_contact_offset	Optional target-support-frame override for the contact's planar offset.
tool_calibrations	Per-control-part tool-frame overrides for asymmetric robot arms.

approach_height: float

Distance above the contact pose used for the free-space approach.

completion_tolerance: float

Planar target distance at which the action succeeds without moving.

contact_distance: float

Initial planar clearance behind the object along the push direction.

contact_frame_to_eef: Tensor

Contact-frame to robot end-effector SE(3) calibration.

hand_interp_steps: int

Number of waypoints used to close the end effector before approach.

object_contact_offset: Tensor

Object-local point used as the center of the planar contact frame.

push_overshoot: float

Additional end-effector travel beyond the object’s target displacement.

retract_height: float

Distance above the pushed pose used for the final retraction.

support_frame_planar_contact_offset: Tensor | None

Optional target-support-frame override for the contact’s planar offset.

tool_calibrations: tuple[PushObjectToolCalibration, ...]

Per-control-part tool-frame overrides for asymmetric robot arms.

class embodichain.lab.sim.atomic_actions.primitives.push_object.PushObjectToolCalibration

Bases: object

End-effector calibration selected by a bound motion control part.

Parameters:

• control_part (str) – Exact motion control-part identifier that selects this calibration.

• contact_frame_to_eef (Tensor) – Contact-frame to robot end-effector SE(3) calibration.

• contact_distance (float | None) – Optional tool-specific planar clearance behind the object. None uses PushObjectOptions’ default.

Attributes:

contact_distance	Optional tool-specific planar clearance behind the object.
contact_frame_to_eef	Contact-frame to robot end-effector SE(3) calibration.
control_part	Exact motion control-part identifier that selects this calibration.

contact_distance: float | None

Optional tool-specific planar clearance behind the object.

contact_frame_to_eef: Tensor

Contact-frame to robot end-effector SE(3) calibration.

control_part: str

Exact motion control-part identifier that selects this calibration.

Place

Place atomic action implementation.

Classes:

AssembleGoal	Place a held assemble object onto a base object at a relative pose.
Place	Lower the held object to a place pose, open the gripper, retract.
PlaceGoal	End-effector release-pose target used by Place.
PlaceOptions	Per-invocation placement behavior.

class embodichain.lab.sim.atomic_actions.primitives.place.AssembleGoal

Bases: object

Place a held assemble object onto a base object at a relative pose.

The base object pose is a late-bound SceneEntityPose. The assemble object’s target pose is base_pose @ assemble_to_base_pose. The held-object transform (object_to_eef) is read from PlanningContext for the place control part, which a prior PickUp populates.

Attributes:

affordance	Assembly affordance anchoring the assemble object to the base object.
base_pose	Late-bound base-object pose used for snapshot-consistent planning.

affordance: AssembleAffordance

Assembly affordance anchoring the assemble object to the base object.

base_pose: SceneEntityPose

Late-bound base-object pose used for snapshot-consistent planning.

class embodichain.lab.sim.atomic_actions.primitives.place.Place

Bases: AtomicAction[PlaceGoal | AssembleGoal, PlaceOptions]

Lower the held object to a place pose, open the gripper, retract.

The PlaceGoal may carry either a single waypoint (num_envs, 4, 4) (or a broadcastable (4, 4)) or a multi-waypoint trajectory (num_envs, n_waypoint, 4, 4). In the multi-waypoint case the approach segment visits every waypoint in order; approaching from above the first waypoint, descending through each waypoint, then opening the gripper at the final waypoint and retracting to above the last waypoint. Starting joint positions are inherited from PlanningContext.

An AssembleGoal replaces the explicit EEF pose with an assembly affordance: the place pose is derived from the base object’s snapshot pose and assemble_to_base_pose, converted to an EEF pose through the held object’s object_to_eef (read from PlanningContext).

Attributes:

GoalType	Concrete goal dataclass or dataclasses accepted by this skill.
binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

Classes:

OptionsType

alias of PlaceOptions

GoalType: ClassVar[type | tuple[type, ...]] = (<class 'embodichain.lab.sim.atomic_actions.primitives.place.PlaceGoal'>, <class 'embodichain.lab.sim.atomic_actions.primitives.place.AssembleGoal'>)

Concrete goal dataclass or dataclasses accepted by this skill.

OptionsType

alias of PlaceOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'place'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.place.PlaceGoal

Bases: object

End-effector release-pose target used by Place.

Attributes:

tcp_symmetry	Optional TCP-frame symmetry allowed by the placement semantics.
xpos	Target end-effector release pose.

tcp_symmetry: Literal['none', 'z_roll_180']

Optional TCP-frame symmetry allowed by the placement semantics.

"none" preserves the pose exactly. "z_roll_180" lets placement choose between the pose and its TCP z-roll 180 equivalent, which flips TCP x/y while preserving TCP z and translation.

xpos: Tensor | SceneEntityPose

Target end-effector release pose.

Accepts (4, 4), (num_envs, 4, 4), or (num_envs, n_waypoint, 4, 4).

class embodichain.lab.sim.atomic_actions.primitives.place.PlaceOptions

Bases: ActionOptions

Per-invocation placement behavior.

Attributes:

cartesian_waypoint_count	Number of fixed-orientation Cartesian keyframes per translation segment.
hand_interp_steps	Number of waypoints for the gripper-open interpolation segment.
lift_height	Height (m) to retract the end-effector after opening the gripper.
max_approach_retract_z	Optional maximum world-frame TCP z for approach and retract poses (m).
preserve_current_object_orientation	Keep the held object's observed world orientation at the place target.
release_settle_steps	Fully open hold frames before retracting the end-effector.

cartesian_waypoint_count: int

Number of fixed-orientation Cartesian keyframes per translation segment.

hand_interp_steps: int

Number of waypoints for the gripper-open interpolation segment.

lift_height: float

Height (m) to retract the end-effector after opening the gripper.

max_approach_retract_z: float | None

Optional maximum world-frame TCP z for approach and retract poses (m).

preserve_current_object_orientation: bool

Keep the held object’s observed world orientation at the place target.

release_settle_steps: int

Fully open hold frames before retracting the end-effector.

Press

Press atomic action implementation.

Classes:

Press	Open-loop motion primitive that approaches, presses, and retracts.
PressGoal	Target object described by a press affordance.
PressOptions	Per-invocation pressing behavior.

class embodichain.lab.sim.atomic_actions.primitives.press.Press

Bases: AtomicAction[PressGoal, PressOptions]

Open-loop motion primitive that approaches, presses, and retracts.

Classes:

GoalType	alias of PressGoal
OptionsType	alias of PressOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of PressGoal

OptionsType

alias of PressOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'press'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.press.PressGoal

Bases: ObjectActionGoal

Target object described by a press affordance.

Attributes:

target_pose

Target pose snapshot or late-bound stable scene-entity reference.

target_pose: Tensor | SceneEntityPose

Target pose snapshot or late-bound stable scene-entity reference.

class embodichain.lab.sim.atomic_actions.primitives.press.PressOptions

Bases: ActionOptions

Per-invocation pressing behavior.

Attributes:

approach_distance	Distance from the press position opposite the press direction.
hand_interp_steps	Number of waypoints used to close the hand.
press_distance	Distance traveled into the target along its press axis.
press_position	Optional local-frame position overriding the affordance press position.

approach_distance: float

Distance from the press position opposite the press direction.

hand_interp_steps: int

Number of waypoints used to close the hand.

press_distance: float

Distance traveled into the target along its press axis.

press_position: tuple[float, float, float] | None

Optional local-frame position overriding the affordance press position.

Slide

Slide atomic action implementation.

Classes:

Slide	Open-loop approach, grasp, and axis-constrained sliding motion.
SlideGoal	Translating articulation link described by a slide affordance.
SlideOptions	Per-invocation sliding behavior for a translating articulation link.

class embodichain.lab.sim.atomic_actions.primitives.slide.Slide

Bases: AtomicAction[SlideGoal, SlideOptions]

Open-loop approach, grasp, and axis-constrained sliding motion.

Classes:

GoalType	alias of SlideGoal
OptionsType	alias of SlideOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of SlideGoal

OptionsType

alias of SlideOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'slide'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.slide.SlideGoal

Bases: ObjectActionGoal

Translating articulation link described by a slide affordance.

Attributes:

target_pose

Link pose snapshot or late-bound stable scene-entity reference.

target_pose: Tensor | SceneEntityPose

Link pose snapshot or late-bound stable scene-entity reference.

class embodichain.lab.sim.atomic_actions.primitives.slide.SlideOptions

Bases: ActionOptions

Per-invocation sliding behavior for a translating articulation link.

Attributes:

approach_distance	Pre-grasp distance opposite the approach/push axis.
direction	Whether to pull the part open or push it closed.
hand_interp_steps	Number of waypoints used for each close/open hand segment.
translation_distance	Distance traveled along the pull or push direction.

approach_distance: float

Pre-grasp distance opposite the approach/push axis.

direction: Literal['pull', 'push']

Whether to pull the part open or push it closed.

hand_interp_steps: int

Number of waypoints used for each close/open hand segment.

translation_distance: float

Distance traveled along the pull or push direction.

OpenDoor

OpenDoor atomic action implementation.

Classes:

OpenDoor	Approach, grasp, rotate a door about its hinge, release, and retract.
OpenDoorGoal	Door handle and desired absolute opening state.
OpenDoorOptions	Per-invocation approach, interpolation, release, and retract behavior.

class embodichain.lab.sim.atomic_actions.primitives.open_door.OpenDoor

Bases: AtomicAction[OpenDoorGoal, OpenDoorOptions]

Approach, grasp, rotate a door about its hinge, release, and retract.

Classes:

GoalType	alias of OpenDoorGoal
OptionsType	alias of OpenDoorOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of OpenDoorGoal

OptionsType

alias of OpenDoorOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'open_door'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.open_door.OpenDoorGoal

Bases: ObjectActionGoal

Door handle and desired absolute opening state.

Attributes:

open_fraction	Desired hinge position normalized from its closed to open legal endpoint.
target_pose	Handle-link pose snapshot or late-bound scene-entity reference.

open_fraction: float | Tensor

Desired hinge position normalized from its closed to open legal endpoint.

target_pose: Tensor | SceneEntityPose

Handle-link pose snapshot or late-bound scene-entity reference.

class embodichain.lab.sim.atomic_actions.primitives.open_door.OpenDoorOptions

Bases: ActionOptions

Per-invocation approach, interpolation, release, and retract behavior.

Attributes:

approach_distance	Pre-grasp distance opposite the automatically inferred approach axis.
door_waypoint_count	Number of Cartesian keyframes along the handle's circular arc.
hand_interp_steps	Number of waypoints used for each close/open hand segment.
joint_position_tolerance	Tolerance for legal-limit and already-open comparisons in radians.
retract_distance	Post-release retreat distance opposite the rotated approach axis.

approach_distance: float

Pre-grasp distance opposite the automatically inferred approach axis.

door_waypoint_count: int

Number of Cartesian keyframes along the handle’s circular arc.

hand_interp_steps: int

Number of waypoints used for each close/open hand segment.

joint_position_tolerance: float

Tolerance for legal-limit and already-open comparisons in radians.

retract_distance: float

Post-release retreat distance opposite the rotated approach axis.

Twist

Twist atomic action implementation.

Classes:

Twist	Open-loop approach, grasp, twist, release, and retract motion.
TwistGoal	Target object described by a twist affordance.
TwistOptions	Per-invocation twisting behavior.

class embodichain.lab.sim.atomic_actions.primitives.twist.Twist

Bases: AtomicAction[TwistGoal, TwistOptions]

Open-loop approach, grasp, twist, release, and retract motion.

Classes:

GoalType	alias of TwistGoal
OptionsType	alias of TwistOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of TwistGoal

OptionsType

alias of TwistOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='primary', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)),), constraints=())

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'twist'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.twist.TwistGoal

Bases: ObjectActionGoal

Target object described by a twist affordance.

Attributes:

target_pose

Target pose snapshot or late-bound stable scene-entity reference.

target_pose: Tensor | SceneEntityPose

Target pose snapshot or late-bound stable scene-entity reference.

class embodichain.lab.sim.atomic_actions.primitives.twist.TwistOptions

Bases: ActionOptions

Per-invocation twisting behavior.

Attributes:

hand_interp_steps	Number of waypoints used for each close/open hand segment.
pre_grasp_distance	Distance from the grasp pose along its negative z-axis.
twist_angle	Requested twist rotation in radians.
twist_waypoint_count	Number of Cartesian keyframes along the target's circular twist arc.

hand_interp_steps: int

Number of waypoints used for each close/open hand segment.

pre_grasp_distance: float

Distance from the grasp pose along its negative z-axis.

twist_angle: float

Requested twist rotation in radians.

twist_waypoint_count: int

Number of Cartesian keyframes along the target’s circular twist arc.

CoordinatedPickment

CoordinatedPickment atomic action implementation.

Classes:

CoordinatedPickGoal	Object-centric target for picking and moving one object with two hands.
CoordinatedPickment	Pick and move a single object pinched by two hands.
CoordinatedPickmentOptions	Per-invocation coordinated pickup behavior.

class embodichain.lab.sim.atomic_actions.primitives.coordinated_pickment.CoordinatedPickGoal

Bases: ObjectActionGoal

Object-centric target for picking and moving one object with two hands.

The left/right grasp poses are not supplied by the caller; they are sampled by the parallel-jaw grasp-pose service at planning time using the dual-arm direction and approach direction declared on CoordinatedPickmentOptions.

Attributes:

object_initial_pose	Optional initial object pose.
object_target_pose	Target pose for the shared object, shape (4, 4) or (num_envs, 4, 4).

object_initial_pose: Tensor | SceneEntityPose | None

Optional initial object pose.

When omitted, the pose is grounded through the semantic object’s stable scene identity.

object_target_pose: Tensor | SceneEntityPose

Target pose for the shared object, shape (4, 4) or (num_envs, 4, 4).

class embodichain.lab.sim.atomic_actions.primitives.coordinated_pickment.CoordinatedPickment

Bases: AtomicAction[CoordinatedPickGoal, CoordinatedPickmentOptions]

Pick and move a single object pinched by two hands.

Classes:

GoalType	alias of CoordinatedPickGoal
OptionsType	alias of CoordinatedPickmentOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of CoordinatedPickGoal

OptionsType

alias of CoordinatedPickmentOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='left', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.inverse'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)), SkillResourceSlot(slot_id='right', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.inverse'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),))), constraints=(DisjointResourceSlots(slots=('left', 'right')),))

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'coordinated_pickment'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.coordinated_pickment.CoordinatedPickmentOptions

Bases: ActionOptions

Per-invocation coordinated pickup behavior.

Left/right grasps are sampled from target-local affordance geometry by the engine’s parallel-jaw grasp-pose service. The dual-arm direction splits the object into left/right grasp regions and the approach direction filters the sampled antipodal pairs.

Attributes:

approach_direction	World-frame direction used to sample and approach both grasps, shape (3,).
hand_interp_steps	Number of waypoints used for the simultaneous hand-close segment.
hold_steps	Number of waypoints to hold the final object target pose.
left_to_right_arm_direction	World-frame direction from the left arm base to the right arm base, shape (3,).
lift_height	World-Z lift distance before moving to the object target pose.
middle_empty_ratio	Fraction of the object's left-to-right extent left grasp-free in the middle so the two grippers pinch opposite ends.
object_motion_keyframes	Number of object-pose keyframes solved by IK before joint-space interpolation.
pre_grasp_distance	World distance to retreat from each grasp pose along negative TCP z.

approach_direction: Tensor

World-frame direction used to sample and approach both grasps, shape (3,).

hand_interp_steps: int

Number of waypoints used for the simultaneous hand-close segment.

hold_steps: int

Number of waypoints to hold the final object target pose.

left_to_right_arm_direction: Tensor

World-frame direction from the left arm base to the right arm base, shape (3,). It partitions the object into left/right grasp regions and should be a finite, non-zero vector; it is normalized at planning time.

lift_height: float

World-Z lift distance before moving to the object target pose.

middle_empty_ratio: float

Fraction of the object’s left-to-right extent left grasp-free in the middle so the two grippers pinch opposite ends. Must be in [0, 1].

object_motion_keyframes: int

Number of object-pose keyframes solved by IK before joint-space interpolation.

pre_grasp_distance: float

World distance to retreat from each grasp pose along negative TCP z.

CoordinatedPlacement

CoordinatedPlacement atomic action implementation.

Classes:

CoordinatedPlacement	Coordinate two held objects: support object below, placing object above.
CoordinatedPlacementGoal	Object-centric target for dual-arm coordinated placement.
CoordinatedPlacementOptions	Per-invocation coordinated placement behavior.

class embodichain.lab.sim.atomic_actions.primitives.coordinated_placement.CoordinatedPlacement

Bases: AtomicAction[CoordinatedPlacementGoal, CoordinatedPlacementOptions]

Coordinate two held objects: support object below, placing object above.

Classes:

GoalType	alias of CoordinatedPlacementGoal
OptionsType	alias of CoordinatedPlacementOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
skill_id	Stable registry identifier for this skill.

GoalType

alias of CoordinatedPlacementGoal

OptionsType

alias of CoordinatedPlacementOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='placing', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)), SkillResourceSlot(slot_id='support', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),))), constraints=(DisjointResourceSlots(slots=('placing', 'support')),))

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

skill_id: ClassVar[str] = 'coordinated_placement'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.coordinated_placement.CoordinatedPlacementGoal

Bases: object

Object-centric target for dual-arm coordinated placement.

Attributes:

placing_height_offset	World-Z offset above the placing object target pose.
placing_object_target_pose	Target pose for the object released by the placing arm.
release	Whether the placing hand releases.
support_height_offset	World-Z offset above the support object target pose.
support_object_target_pose	Target pose for the object held by the support arm.

placing_height_offset: float | None

World-Z offset above the placing object target pose.

placing_object_target_pose: Tensor | SceneEntityPose

Target pose for the object released by the placing arm.

release: bool | None

Whether the placing hand releases. None uses invocation options.

support_height_offset: float | None

World-Z offset above the support object target pose.

support_object_target_pose: Tensor | SceneEntityPose

Target pose for the object held by the support arm.

class embodichain.lab.sim.atomic_actions.primitives.coordinated_placement.CoordinatedPlacementOptions

Bases: ActionOptions

Per-invocation coordinated placement behavior.

Attributes:

hand_interp_steps	Number of waypoints for the placing-hand release interpolation.
hold_steps	Number of waypoints to hold alignment before releasing.
lift_height	World-Z lift distance for the placing arm after release.
placing_height_offset	Default World-Z offset above the placing object target pose.
release	Whether to open the placing hand at the aligned placement pose.
retreat_steps	Number of waypoints used for the placing-arm lift retreat.
support_height_offset	Default World-Z offset above the support object target pose.

hand_interp_steps: int

Number of waypoints for the placing-hand release interpolation.

hold_steps: int

Number of waypoints to hold alignment before releasing.

lift_height: float

World-Z lift distance for the placing arm after release.

placing_height_offset: float

Default World-Z offset above the placing object target pose.

release: bool

Whether to open the placing hand at the aligned placement pose.

retreat_steps: int

Number of waypoints used for the placing-arm lift retreat.

support_height_offset: float

Default World-Z offset above the support object target pose.

HandOver

Pick-up-and-handover atomic action implementation.

Classes:

HandOver	Pick an object with the nearer arm, hand it over, and place it.
HandOverGoal	Object to pick and hand over, plus its final object pose.
HandOverOptions	Per-invocation pick-up, handover, and placement behavior.

class embodichain.lab.sim.atomic_actions.primitives.hand_over.HandOver

Bases: AtomicAction[HandOverGoal, HandOverOptions]

Pick an object with the nearer arm, hand it over, and place it.

For each environment, the action chooses the arm whose root link is closer to the observed object pose. It samples at most 1000 mesh-surface points and applies SVD in the current object pose to find obj_longest_axis. When that axis is closer to world Z than to the horizontal plane, both grasp approaches point toward the object horizontally and tilt downward by 45 degrees. Otherwise both approaches are world-Z downward.

The first arm grasps the projected end of obj_longest_axis nearest its current TCP; the receiving arm grasps the opposite end at the predicted middle object pose. This keeps the two hands from selecting the same object region regardless of whether a long object is standing or lying down.

After each grasp waypoint, subsequent EEF waypoints preserve that grasp rotation and change translation only. In particular, placement first moves strictly horizontally at the handover height and then lowers to the final target pose before releasing the object.

Classes:

GoalType	alias of HandOverGoal
OptionsType	alias of HandOverOptions

Attributes:

binding_contract	Explicit robot-independent requirements for semantic discovery.
open_loop	Whether the skill intentionally declares no verified physical effect.
skill_id	Stable registry identifier for this skill.

GoalType

alias of HandOverGoal

OptionsType

alias of HandOverOptions

binding_contract: ClassVar[SkillBindingContract] = SkillBindingContract(slots=(SkillResourceSlot(slot_id='source', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),)), SkillResourceSlot(slot_id='destination', endpoints=(SkillEndpointRequirement(endpoint_id='motion', capabilities=frozenset({'kinematics.forward', 'motion.cartesian_pose'}), required_commands=mappingproxy({})), SkillEndpointRequirement(endpoint_id='grasp', capabilities=frozenset({'interaction.grasp'}), required_commands=mappingproxy({'open': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>, 'grasp': <class 'embodichain.lab.sim.atomic_actions.control.JointPositionCommand'>}))), constraints=(DisjointSlotEndpoints(endpoint_ids=('motion', 'grasp')),))), constraints=(DisjointResourceSlots(slots=('source', 'destination')),))

Explicit robot-independent requirements for semantic discovery.

Concrete action classes must declare this attribute in their own class body to opt into the semantic catalog. Inheriting another action’s contract does not silently expose a new skill identifier.

open_loop: ClassVar[bool] = True

Whether the skill intentionally declares no verified physical effect.

skill_id: ClassVar[str] = 'hand_over'

Stable registry identifier for this skill.

class embodichain.lab.sim.atomic_actions.primitives.hand_over.HandOverGoal

Bases: ObjectActionGoal

Object to pick and hand over, plus its final object pose.

Attributes:

target_pose

Final object pose after the receiving arm lowers and releases it.

target_pose: PoseGoalValue

Final object pose after the receiving arm lowers and releases it.

class embodichain.lab.sim.atomic_actions.primitives.hand_over.HandOverOptions

Bases: ActionOptions

Per-invocation pick-up, handover, and placement behavior.

Attributes:

hand_interp_steps	Waypoints used by every gripper open/close interpolation.
lift_height	World-Z distance used to lift the object after the first grasp.
pre_grasp_distance	Distance from each grasp pose to its approach pose, in metres.

hand_interp_steps: int

Waypoints used by every gripper open/close interpolation.

lift_height: float

World-Z distance used to lift the object after the first grasp.

pre_grasp_distance: float

Distance from each grasp pose to its approach pose, in metres.