embodichain.lab.visualization

Browser-based visualization of simulation scenes via Viser.

SceneExporter produces backend-neutral snapshots, VisualizationRuntime pushes them to a Viser backend, with gizmo overlays, camera preview, and CLI helpers.

Overview

Browser-based visualization of simulation scenes. The SceneExporter reads simulation assets on the simulation thread and produces detached, backend-neutral CPU snapshots (SceneManifest, SceneFrame, CameraImageFrame). A background VisualizationRuntime owns latest-frame queues, rate limiting, health, and telemetry, and pushes snapshots to a visualization backend - currently the Viser server (ViserServerCfg) that publishes an interactive 3D view to the browser. The stack supports interactive gizmos and scalar articulation joint controls (with optional command write-back to the simulation), scene overlays (targets, trajectories, point clouds), and live RGB camera preview. CLI helpers (add_viser_args_to_parser(), visualization_cfg_from_args()) wire the standard --viser* arguments into launchers.

Configuration

class embodichain.lab.visualization.VisualizationCfg

Bases: object

Configure live scene visualization.

Parameters:

• backend (str) – Visualization backend name. Supported values are "none" and "viser".

• scene_fps (float) – Maximum scene capture rate.

• env_ids (list[int] | None) – Environment indices exposed by the visualizer. None selects every simulation environment.

• max_visible_envs (int | None) – Optional safety limit on the number of selected environments. None disables the limit.

• point_cloud_max_points (int) – Maximum number of points retained per point cloud.

• sensor_image_fps (float | None) – Maximum camera RGB preview update rate. None captures once per visualization step instead of using wall-clock rate limiting.

• soft_body_fps (float) – Maximum soft-body and cloth vertex update rate.

• allow_commands (bool) – Whether simulation-mutating browser commands are allowed. This enables Viser Gizmo dragging and registered articulation joint controls. Keep it disabled for untrusted or publicly reachable browser sessions.

• viser_server (ViserServerCfg) – Viser HTTP/WebSocket server binding settings.

Attributes:

allow_commands
backend
env_ids
max_visible_envs
point_cloud_max_points
scene_fps
sensor_image_fps
soft_body_fps
viser_server

allow_commands: bool

backend: str

env_ids: list[int] | None

max_visible_envs: int | None

point_cloud_max_points: int

scene_fps: float

sensor_image_fps: float | None

soft_body_fps: float

viser_server: ViserServerCfg

class embodichain.lab.visualization.ViserServerCfg

Bases: object

Configure the local Viser HTTP/WebSocket server.

Parameters:

• host (str) – Interface on which Viser listens. Server deployments should inject this value instead of accepting it from an untrusted run request.

• port (int) – TCP port allocated to the worker.

• label (str) – Browser application label.

• verbose (bool) – Whether Viser should print server diagnostics.

Attributes:

host
label
port
verbose

host: str

label: str

port: int

verbose: bool

Runtime

class embodichain.lab.visualization.VisualizationRuntime

Bases: object

Run scene capture and a visualization backend without blocking simulation.

The simulation thread calls capture(). Viser creation and all handle mutations occur on one private update thread. The frame queue always keeps the newest sample, preventing visualization overload from accumulating lag.

Parameters:

• exporter (SceneExporter) – Scene exporter bound to a simulation manager.

• cfg (VisualizationCfg) – Visualization, frame rate, and Viser server configuration.

• backend (VisualizationBackend | None) – Optional backend injection hook used by tests and alternate UIs.

Methods:

__init__(exporter, cfg[, backend])
capture(*, sim_step, sim_time[, overlays, ...])	Capture a due frame and enqueue it without waiting for Viser.
drain_gizmo_commands()	Drain browser Gizmo commands for simulation-thread processing.
drain_joint_control_commands()	Drain browser joint commands for simulation-thread processing.
drain_pick_commands()	Drain browser click-pick commands for simulation-thread processing.
drain_replay_control_command()	Return the newest browser replay seek, if one is pending.
publish_replay_control(*, step, max_step[, ...])	Asynchronously publish trajectory replay progress to Viser.
refresh_scene()	Capture and asynchronously publish a new topology revision.
set_joint_control_provider(provider)	Install a simulation-thread joint source for future scene captures.
start([timeout])	Build the initial manifest and start the backend update thread.
stop([timeout])	Stop the backend and reclaim its server port and update thread.

Attributes:

endpoint	Local browser endpoint after start() returns.
health	Return runtime, endpoint, client, and worker health information.
is_running	Whether the visualization update thread is alive.
stats	Return an immutable telemetry snapshot.

__init__(exporter, cfg, backend=None)

capture(*, sim_step, sim_time, overlays=None, force=False, capture_camera_images=True)

Capture a due frame and enqueue it without waiting for Viser.

Parameters:

• sim_step (int) – Current simulation step.

• sim_time (float) – Current simulation time in seconds.

• overlays (SceneOverlays | None) – Optional backend-neutral debug overlays.

• force (bool) – Ignore the configured scene FPS limiter.

• capture_camera_images (bool) – Whether camera images may be captured in this call. Simulation batches disable this for intermediate physics substeps.

Return type:

bool

Returns:

True when a frame was captured, otherwise False when limited.

drain_gizmo_commands()

Drain browser Gizmo commands for simulation-thread processing.

Return type:

tuple[GizmoCommand, ...]

drain_joint_control_commands()

Drain browser joint commands for simulation-thread processing.

Return type:

tuple[JointControlCommand, ...]

drain_pick_commands()

Drain browser click-pick commands for simulation-thread processing.

Return type:

tuple[PickCommand, ...]

drain_replay_control_command()

Return the newest browser replay seek, if one is pending.

Return type:

int | None

Returns:

Requested trajectory step, or None when no seek is pending.

property endpoint: str | None

Local browser endpoint after start() returns.

property health: RuntimeHealth

Return runtime, endpoint, client, and worker health information.

property is_running: bool

Whether the visualization update thread is alive.

publish_replay_control(*, step, max_step, visible=True)

Asynchronously publish trajectory replay progress to Viser.

Parameters:

• step (int) – Current trajectory step.

• max_step (int) – Largest valid trajectory step.

• visible (bool) – Whether the replay control should be visible.

Raises:

• RuntimeError – If the visualization runtime is not running.

• ValueError – If the step range is invalid.

Return type:

None

refresh_scene()

Capture and asynchronously publish a new topology revision.

Return type:

SceneManifest

set_joint_control_provider(provider)

Install a simulation-thread joint source for future scene captures.

Registering a provider does not publish a new manifest by itself. The caller must refresh the scene after registration so the backend can build its controls.

Return type:

None

start(timeout=10.0)

Build the initial manifest and start the backend update thread.

Return type:

None

property stats: RuntimeStats

Return an immutable telemetry snapshot.

stop(timeout=10.0)

Stop the backend and reclaim its server port and update thread.

Return type:

None

class embodichain.lab.visualization.LatestFrameQueue

A one-slot queue where producers replace an unconsumed old frame.

Methods:

__init__()
clear()	Discard any queued frame.
get([timeout])	Return the queued frame, waiting for up to timeout seconds.
get_nowait()	Return the queued frame without blocking.
put_latest(frame)	Enqueue frame and return whether an older frame was dropped.

__init__()

clear()

Discard any queued frame.

Return type:

None

get(timeout=None)

Return the queued frame, waiting for up to timeout seconds.

Return type:

TypeVar(FrameT)

get_nowait()

Return the queued frame without blocking.

Return type:

TypeVar(FrameT)

put_latest(frame)

Enqueue frame and return whether an older frame was dropped.

Return type:

bool

class embodichain.lab.visualization.GizmoCommandQueue

Bounded command queue that coalesces high-rate drag updates.

Drag lifecycle commands are retained. When the queue reaches its soft capacity, an older update for the same Gizmo/client is replaced first.

Methods:

__init__([maxsize])
clear()	Discard all queued commands.
drain()	Return and clear all queued commands in arrival order.
put(command)	Enqueue a command without blocking the Viser callback thread.

__init__(maxsize=256)

clear()

Discard all queued commands.

Return type:

None

drain()

Return and clear all queued commands in arrival order.

Return type:

tuple[GizmoCommand, ...]

put(command)

Enqueue a command without blocking the Viser callback thread.

Return type:

None

class embodichain.lab.visualization.JointControlCommandQueue

Bounded queue that keeps only the newest value for each joint control.

Methods:

__init__([maxsize])
clear()	Discard all queued commands.
drain()	Return and clear all queued commands in arrival order.
put(command)	Enqueue a value without blocking the Viser callback thread.

__init__(maxsize=256)

clear()

Discard all queued commands.

Return type:

None

drain()

Return and clear all queued commands in arrival order.

Return type:

tuple[JointControlCommand, ...]

put(command)

Enqueue a value without blocking the Viser callback thread.

Return type:

None

class embodichain.lab.visualization.RuntimeHealth

Current visualization runtime health and connection state.

Methods:

__init__(status, running, endpoint, ...[, ...])

Attributes:

client_count
endpoint
published_scene_revision
running
status
worker_error

__init__(status, running, endpoint, client_count, published_scene_revision, worker_error=None)

client_count: int

endpoint: str | None

published_scene_revision: int

running: bool

status: str

worker_error: str | None = None

class embodichain.lab.visualization.RuntimeStats

Snapshot of scene and camera-image capture/upload telemetry.

Methods:

__init__([captured_frames, ...])

Attributes:

capture_seconds
captured_frames
captured_image_frames
dropped_frames
dropped_image_frames
frame_bytes
image_bytes
image_capture_seconds
image_upload_seconds
manifest_bytes
published_frames
published_image_frames
rejected_frames
rejected_image_frames
upload_seconds

__init__(captured_frames=0, published_frames=0, dropped_frames=0, rejected_frames=0, manifest_bytes=0, frame_bytes=0, capture_seconds=0.0, upload_seconds=0.0, captured_image_frames=0, published_image_frames=0, dropped_image_frames=0, rejected_image_frames=0, image_bytes=0, image_capture_seconds=0.0, image_upload_seconds=0.0)

capture_seconds: float = 0.0

captured_frames: int = 0

captured_image_frames: int = 0

dropped_frames: int = 0

dropped_image_frames: int = 0

frame_bytes: int = 0

image_bytes: int = 0

image_capture_seconds: float = 0.0

image_upload_seconds: float = 0.0

manifest_bytes: int = 0

published_frames: int = 0

published_image_frames: int = 0

rejected_frames: int = 0

rejected_image_frames: int = 0

upload_seconds: float = 0.0

Scene Export

class embodichain.lab.visualization.SceneExporter

Bases: object

Capture backend-neutral scene data from a SimulationManager.

Scene access happens only in build_manifest() and capture(), so callers can keep both methods on the simulation thread. Returned protocol objects own detached CPU copies and are safe to hand to a background backend.

Parameters:

• sim (SimulationManager) – Simulation manager whose public asset interfaces are exported.

• cfg (VisualizationCfg) – Visualization selection and sampling configuration.

• run_id (str | None) – Stable identifier for this runtime. A UUID is generated by default.

Methods:

__init__(sim, cfg[, run_id])
build_manifest()	Capture scene topology, incrementing the scene revision.
capture(*, sim_step, sim_time[, overlays, ...])	Capture one dynamic scene frame on the simulation thread.
capture_camera_images(*, sim_step, sim_time)	Render and detach one RGB image for every exported camera instance.
resolve_node_target(node_id)	Map a published scene node id to its (uid, kind).
set_joint_control_provider(provider)	Install an optional simulation-thread joint-control source.

Attributes:

has_cameras	Whether the current manifest contains an RGB-capable camera.
has_deformables	Whether the current manifest contains soft-body or cloth nodes.
scene_revision	Current manifest revision, or zero before the first manifest.

__init__(sim, cfg, run_id=None)

build_manifest()

Capture scene topology, incrementing the scene revision.

Call this after assets are created and again after topology-changing resets. Empty articulation links are ignored.

Return type:

SceneManifest

capture(*, sim_step, sim_time, overlays=None, capture_dynamic_geometry=True)

Capture one dynamic scene frame on the simulation thread.

Parameters:

• sim_step (int) – Current simulation step.

• sim_time (float) – Current simulation time in seconds.

• overlays (SceneOverlays | None) – Optional debug overlays.

• capture_dynamic_geometry (bool) – Whether to copy soft-body and cloth vertices.

Return type:

CaptureResult

Returns:

Captured frame and producer-side copy duration.

capture_camera_images(*, sim_step, sim_time)

Render and detach one RGB image for every exported camera instance.

Return type:

CameraImageCaptureResult

property has_cameras: bool

Whether the current manifest contains an RGB-capable camera.

property has_deformables: bool

Whether the current manifest contains soft-body or cloth nodes.

resolve_node_target(node_id)

Map a published scene node id to its (uid, kind).

Used by the simulation thread to turn a Viser click-pick result into the asset uid that SimulationManager.enable_gizmo() expects.

Parameters:

node_id (str) – Scene node id from the current manifest.

Return type:

tuple[str, str] | None

Returns:

(uid, kind) where kind is the asset kind (for example "rigid", "robot", or "articulation"), or None if the node id is not part of the current scene.

property scene_revision: int

Current manifest revision, or zero before the first manifest.

set_joint_control_provider(provider)

Install an optional simulation-thread joint-control source.

The next build_manifest() call snapshots the provider’s static controls. Dynamic values are then sampled by capture().

Return type:

None

class embodichain.lab.visualization.SceneManifest

Static scene topology and geometry for one scene revision.

Methods:

__init__(run_id, scene_revision, nodes, ...)

Attributes:

cameras
geometries
gizmos
joint_controls
length_unit
nodes
run_id
scene_revision
schema_version
up_direction

__init__(run_id, scene_revision, nodes, geometries, cameras=<factory>, gizmos=<factory>, joint_controls=<factory>, schema_version=5, up_direction='+z', length_unit='meter')

cameras: tuple[CameraSpec, ...]

geometries: tuple[MeshGeometry, ...]

gizmos: tuple[GizmoSpec, ...]

joint_controls: tuple[JointControlSpec, ...]

length_unit: str = 'meter'

nodes: tuple[SceneNode, ...]

run_id: str

scene_revision: int

schema_version: int = 5

up_direction: str = '+z'

class embodichain.lab.visualization.SceneFrame

Dynamic poses and overlays for one simulation sample.

Methods:

__init__(run_id, scene_revision, sequence, ...)

Attributes:

camera_ids
camera_positions
camera_wxyz
dynamic_meshes
gizmos
joint_controls
node_ids
overlays
positions
run_id
scene_revision
schema_version
sequence
sim_step
sim_time
visible
wall_time
wxyz

__init__(run_id, scene_revision, sequence, sim_step, sim_time, node_ids, positions, wxyz, visible, camera_ids=<factory>, camera_positions=<factory>, camera_wxyz=<factory>, dynamic_meshes=<factory>, gizmos=<factory>, joint_controls=<factory>, overlays=<factory>, wall_time=<factory>, schema_version=5)

camera_ids: tuple[str, ...]

camera_positions: ndarray

camera_wxyz: ndarray

dynamic_meshes: tuple[DynamicMeshUpdate, ...]

gizmos: tuple[GizmoState, ...]

joint_controls: tuple[JointControlState, ...]

node_ids: tuple[str, ...]

overlays: SceneOverlays

positions: ndarray

run_id: str

scene_revision: int

schema_version: int = 5

sequence: int

sim_step: int

sim_time: float

visible: ndarray

wall_time: str

wxyz: ndarray

class embodichain.lab.visualization.SceneNode

One mesh-bearing logical node in a scene manifest.

Methods:

__init__(node_id, path, parent_id, env_id, ...)

Attributes:

dynamic_geometry
env_id
geometry_id
kind
node_id
parent_id
path
visible

__init__(node_id, path, parent_id, env_id, kind, geometry_id, dynamic_geometry=False, visible=True)

dynamic_geometry: bool = False

env_id: int

geometry_id: str

kind: str

node_id: str

parent_id: str | None

path: str

visible: bool = True

class embodichain.lab.visualization.MeshGeometry

Backend-neutral triangle mesh stored in local coordinates.

Methods:

__init__(geometry_id, vertices, faces[, color])

Attributes:

color
faces
geometry_id
vertices

__init__(geometry_id, vertices, faces, color=(90, 200, 255))

color: tuple[int, int, int] = (90, 200, 255)

faces: ndarray

geometry_id: str

vertices: ndarray

class embodichain.lab.visualization.DynamicMeshUpdate

Detached vertex positions for one deformable scene node.

Methods:

__init__(node_id, vertices)

Attributes:

node_id
vertices

__init__(node_id, vertices)

node_id: str

vertices: ndarray

class embodichain.lab.visualization.GizmoSpec

Static description of one simulation Gizmo exposed by a backend.

Methods:

__init__(gizmo_id, target_uid, target_type, ...)

Attributes:

control_part
env_id
gizmo_id
line_width
path
scale
target_type
target_uid
visible

__init__(gizmo_id, target_uid, target_type, control_part, env_id, path, scale=0.2, line_width=2.5, visible=True)

control_part: str | None

env_id: int

gizmo_id: str

line_width: float = 2.5

path: str

scale: float = 0.2

target_type: str

target_uid: str

visible: bool = True

class embodichain.lab.visualization.GizmoState

Authoritative world pose and visibility of one simulation Gizmo.

Methods:

__init__(gizmo_id, position, wxyz[, visible])

Attributes:

gizmo_id
position
visible
wxyz

__init__(gizmo_id, position, wxyz, visible=True)

gizmo_id: str

position: ndarray

visible: bool = True

wxyz: ndarray

class embodichain.lab.visualization.GizmoCommand

Immutable browser drag command consumed on the simulation thread.

Methods:

__init__(run_id, scene_revision, sequence, ...)

Attributes:

client_id
gizmo_id
phase
position
run_id
scene_revision
schema_version
sequence
wxyz

__init__(run_id, scene_revision, sequence, gizmo_id, phase, client_id, position, wxyz, schema_version=5)

client_id: str

gizmo_id: str

phase: Literal['start', 'update', 'end']

position: ndarray

run_id: str

scene_revision: int

schema_version: int = 5

sequence: int

wxyz: ndarray

PickCommand identifies a node in a specific simulation run and scene revision. An empty selection releases only the gizmo created by click picking; explicitly configured gizmos retain their ownership.

class embodichain.lab.visualization.PickCommand

Immutable browser click-pick command consumed on the simulation thread.

A non-empty node_id requests a Gizmo on the clicked scene node; a None node_id (clicking empty space) clears the picker-owned Gizmo.

Methods:

__init__(run_id, scene_revision, client_id, ...)

Attributes:

client_id
node_id
run_id
scene_revision
schema_version

__init__(run_id, scene_revision, client_id, node_id, schema_version=5)

client_id: str

node_id: str | None

run_id: str

scene_revision: int

schema_version: int = 5

class embodichain.lab.visualization.JointControlSpec

Static description of one scalar articulation joint control.

Values use simulation units: radians for rotational joints and meters for prismatic joints. Controls with two finite limits can be rendered as a slider; controls with one or both limits missing use a numeric input.

Methods:

__init__(control_id, articulation_uid, ...)

Attributes:

articulation_uid
control_id
env_id
initial_value
joint_id
joint_name
joint_type
lower
step
upper

__init__(control_id, articulation_uid, env_id, joint_id, joint_name, joint_type, lower, upper, step, initial_value)

articulation_uid: str

control_id: str

env_id: int

initial_value: float

joint_id: int

joint_name: str

joint_type: Literal['revolute', 'continuous', 'prismatic']

lower: float | None

step: float

upper: float | None

class embodichain.lab.visualization.JointControlState

Authoritative value and command acknowledgement for one joint control.

Methods:

__init__(control_id, value[, applied_sequence])

Attributes:

applied_sequence
control_id
value

__init__(control_id, value, applied_sequence=0)

applied_sequence: int = 0

control_id: str

value: float

class embodichain.lab.visualization.JointControlCommand

Immutable browser joint command consumed on the simulation thread.

Methods:

__init__(run_id, scene_revision, sequence, ...)

Attributes:

client_id
control_id
run_id
scene_revision
schema_version
sequence
value

__init__(run_id, scene_revision, sequence, client_id, control_id, value, schema_version=5)

client_id: str

control_id: str

run_id: str

scene_revision: int

schema_version: int = 5

sequence: int

value: float

class embodichain.lab.visualization.JointControlProvider

Simulation-thread source of optional articulation joint controls.

Methods:

__init__(*args, **kwargs)
joint_control_specs()	Return the static controls to include in the next manifest.
joint_control_states()	Return current values ordered independently of backend state.

__init__(*args, **kwargs)

joint_control_specs()

Return the static controls to include in the next manifest.

Return type:

tuple[JointControlSpec, ...]

joint_control_states()

Return current values ordered independently of backend state.

Return type:

tuple[JointControlState, ...]

class embodichain.lab.visualization.CaptureResult

A captured frame and the time spent copying scene data.

Methods:

__init__(frame, capture_seconds)

Attributes:

capture_seconds
frame

__init__(frame, capture_seconds)

capture_seconds: float

frame: SceneFrame

Overlays and Camera Preview

class embodichain.lab.visualization.SceneOverlays

All optional overlays attached to a dynamic frame.

Methods:

__init__([frames, trajectories, targets, ...])

Attributes:

frames
point_clouds
targets
trajectories

__init__(frames=<factory>, trajectories=<factory>, targets=<factory>, point_clouds=<factory>)

frames: tuple[FrameOverlay, ...]

point_clouds: tuple[PointCloudOverlay, ...]

targets: tuple[TargetOverlay, ...]

trajectories: tuple[TrajectoryOverlay, ...]

class embodichain.lab.visualization.FrameOverlay

Coordinate frame overlay.

Methods:

__init__(overlay_id, position, wxyz[, ...])

Attributes:

axes_length
axes_radius
overlay_id
position
visible
wxyz

__init__(overlay_id, position, wxyz, axes_length=0.15, axes_radius=0.006, visible=True)

axes_length: float = 0.15

axes_radius: float = 0.006

overlay_id: str

position: ndarray

visible: bool = True

wxyz: ndarray

class embodichain.lab.visualization.TargetOverlay

Target pose overlay rendered as a coordinate frame.

Methods:

__init__(overlay_id, position, wxyz[, ...])

Attributes:

axes_length
overlay_id
position
visible
wxyz

__init__(overlay_id, position, wxyz, axes_length=0.2, visible=True)

axes_length: float = 0.2

overlay_id: str

position: ndarray

visible: bool = True

wxyz: ndarray

class embodichain.lab.visualization.TrajectoryOverlay

Polyline trajectory overlay.

Methods:

__init__(overlay_id, points[, color, ...])

Attributes:

color
line_width
overlay_id
points
visible

__init__(overlay_id, points, color=(255, 170, 30), line_width=3.0, visible=True)

color: tuple[int, int, int] = (255, 170, 30)

line_width: float = 3.0

overlay_id: str

points: ndarray

visible: bool = True

class embodichain.lab.visualization.PointCloudOverlay

Point cloud overlay with per-cloud or per-point RGB colors.

Methods:

__init__(overlay_id, points[, colors, ...])

Attributes:

colors
overlay_id
point_size
points
visible

__init__(overlay_id, points, colors=(90, 200, 255), point_size=0.01, visible=True)

colors: ndarray | tuple[int, int, int] = (90, 200, 255)

overlay_id: str

point_size: float = 0.01

points: ndarray

visible: bool = True

class embodichain.lab.visualization.CameraSpec

Static pinhole-camera parameters for one environment instance.

Methods:

__init__(camera_id, sensor_uid, env_id, ...)

Attributes:

aspect
camera_id
env_id
far
fov_y
near
path
role
sensor_uid

__init__(camera_id, sensor_uid, env_id, path, fov_y, aspect, near, far, role='sensor')

aspect: float

camera_id: str

env_id: int

far: float

fov_y: float

near: float

path: str

role: Literal['sensor', 'record'] = 'sensor'

sensor_uid: str

class embodichain.lab.visualization.CameraImage

One detached RGB image associated with a manifest camera.

Methods:

__init__(camera_id, image)

Attributes:

camera_id
image

__init__(camera_id, image)

camera_id: str

image: ndarray

class embodichain.lab.visualization.CameraImageFrame

Low-frequency RGB images captured at one simulation timestamp.

Methods:

__init__(run_id, scene_revision, sequence, ...)

Attributes:

images
run_id
scene_revision
schema_version
sequence
sim_step
sim_time
wall_time

__init__(run_id, scene_revision, sequence, sim_step, sim_time, images, wall_time=<factory>, schema_version=5)

images: tuple[CameraImage, ...]

run_id: str

scene_revision: int

schema_version: int = 5

sequence: int

sim_step: int

sim_time: float

wall_time: str

class embodichain.lab.visualization.CameraImageCaptureResult

Captured RGB images and the time spent rendering and copying them.

Methods:

__init__(frame, capture_seconds)

Attributes:

capture_seconds
frame

__init__(frame, capture_seconds)

capture_seconds: float

frame: CameraImageFrame

Pose Conversion

embodichain.lab.visualization.pose_to_position_wxyz(pose)

Split pose arrays into positions and normalized wxyz quaternions.

The accepted layouts are (..., 7) in EmbodiChain’s (x, y, z, qw, qx, qy, qz) convention or homogeneous (..., 4, 4) matrices. This is the single conversion boundary used by scene exporters.

Parameters:

pose (object) – Pose or batch of poses.

Return type:

tuple[ndarray, ndarray]

Returns:

A pair of float32 arrays containing positions and wxyz quaternions.

Raises:

ValueError – If the shape is unsupported or a quaternion is degenerate.

Command-Line Helpers

embodichain.lab.visualization.add_viser_args_to_parser(parser)

Add the standard EmbodiChain Viser command-line options.

Parameters:

parser (ArgumentParser) – Parser receiving the Viser options.

Return type:

None

embodichain.lab.visualization.visualization_cfg_from_args(args)

Build visualization configuration from parsed CLI arguments.

Parameters:

args (Namespace) – Namespace populated by add_viser_args_to_parser().

Return type:

VisualizationCfg

Returns:

Visualization configuration including Viser server settings.