Preview a Robot

Preview a Robot#

embodichain preview-robot loads one catalog robot or a standalone robot configuration. The default frontend is the native simulation window. python -m embodichain preview-robot is equivalent.

Catalog selection#

embodichain list-robot
embodichain preview-robot --robot FrankaPanda
embodichain preview-robot --robot URRobot --variant ur5e
embodichain preview-robot --robot DualArmRobot --variant ur5 --viser
embodichain preview-robot --robot DexforceW1 --variant v025/DH_PGC --viser

IDs are exact catalog names, including installed third-party providers. --variant uses the variant name displayed by list-robot. Omitting it preserves the robot’s existing construction defaults. --robot-params accepts a JSON object of robot configuration overrides:

embodichain preview-robot --robot DexforceW1 --variant v025/BrainCo \
    --robot-params '{"uid":"inspection_robot"}' --viser

The precedence is catalog defaults, explicit variant defaults, file fields, then CLI robot parameters. These are top-level field overrides; each configuration class retains its existing nested decoding rules.

Robot configuration files#

--robot-config accepts JSON, YAML, or YML and is mutually exclusive with --robot. The file root contains RobotCfg fields:

uid: custom_arm
fpath: ./arm.urdf
root_props:
  fixed_base: true
control_parts:
  arm: ["joint[1-6]"]
solver_cfg:
  arm:
    class_type: PytorchSolver
    root_link_name: base_link
    end_link_name: tool0

Joint and link names must match the chosen asset. Local asset paths resolve relative to the configuration directory. Registered asset references continue to use the shared asset registry. URDF assembly components and solver URDF paths support the same local-path resolution.

An optional top-level class_type selects a catalog configuration rather than generic RobotCfg:

class_type: URRobot
uid: custom_ur
robot_type: ur5

embodichain preview-robot --robot-config /path/to/robot.yaml --viser

--variant requires a catalog selection, either --robot or file class_type. Files contain robot configuration fields; Gym deployments and embodiment components keep their existing loaders.

Interaction#

In the native window, press I to activate or toggle the existing managed IK Gizmos, then drag a control part’s target. In Viser, drag its Gizmo directly. Only defined parts with complete IK chain/TCP metadata are registered, and Gizmo control writes the part’s independent joint targets. Registration does not activate native IK or change the robot’s targets. Automatic Gizmos retain the current single-environment requirement and default DexSim IK backend.

With --viser, the existing joint panel exposes independent scalar joints. It supports joint limits, numeric inputs for continuous/unbounded joints, and reset to the initial pose. Parts without IK metadata remain controllable through the joint panel. Slider commands and Gizmo writes are processed on the simulation thread; subsequent steps adopt the latest drive targets so an IK pose is not overwritten by an old slider pose. When both arrive in one step, Gizmo processing follows slider processing.

# Browser joint controls without IK Gizmos
embodichain preview-robot --robot ur5 --viser --no-gizmo

# Gizmos without the joint panel
embodichain preview-robot --robot ur5 --viser --no-joint-control

The native frontend and Viser are mutually exclusive. Native preview requires a working display and renderer. Common --viser-host, --viser-port, --physics, --device, and renderer arguments are available in preview-robot --help.

Headless checks and exit#

# Load and validate, then exit without stepping
embodichain preview-robot --robot ur5 --headless --renderer no-render

# Load and execute ten physics steps with Newton
embodichain preview-robot --robot ur5 --physics newton \
    --headless --renderer no-render --max-steps 10

Pure headless mode exits after loading unless --max-steps is supplied. Native and Viser previews continue until Ctrl+C or their step limit. Every run owns and cleans up its simulation; existing robot-specific module demonstrations remain available.

Contents