add-atomic-action

To add a custom atomic action to an EmbodiChain environment, define an ActionCfg extending the base config with a unique name, implement the AtomicAction interface with execute() and validate() methods, then register it with AtomicActionEngine and export it in the public API.

The trajectory data returned by an EmbodiChain AtomicAction execute method must be shaped as (n_envs, n_waypoints, len(joint_ids)). The execute() method must return this trajectory array alongside joint_ids and a boolean is_success flag.

A unique name in your ActionCfg is required for registry discovery within the EmbodiChain framework. This registration ensures the custom motion primitive is recognized and can be correctly executed by the AtomicActionEngine across simulation workloads.

Built-in EmbodiChain actions cover standard behaviors, but you need to scaffold new motion primitives when extending AtomicActionEngine with behavior not covered by built-in actions for custom training data generation pipelines.

Yes, EmbodiChain requires a cheap validate() feasibility check in your AtomicAction implementation. This fast validation runs before the execute() method to ensure the trajectory planning action can be performed successfully.

After creating an ActionCfg for embodied AI simulation, you must register the action with AtomicActionEngine, export it from the public package, update supported-actions documentation, and add minimal tests to prevent regressions.