AI Accelerator: Who executes /ur_cumotion/last_planned_trajectory?

ROS
1

Hi,

we are evaluating the UR AI Accelerator with a real UR5e and Isaac Sim.

Current status:

  • ur_cumotion_planner_node is running
  • /IP192168000100/ur_cumotion/plan_motion works
  • /IP192168000100/ur_cumotion/plan_multi_goal_motion works
  • planning returns success=True
  • trajectories are published on
    /IP192168000100/ur_cumotion/last_planned_trajectory

We can successfully generate valid trajectories.

However, we do not understand the intended execution path on the real robot.

Questions:

  1. Which component is supposed to consume the planned JointTrajectory?

  2. Is there a standard execution node provided by the AI Accelerator?

  3. Does trajectory execution happen through:

    • PolyScope X Logic Programs?
    • a ROS2 Action Server?
    • a URCap?
    • another component?

  4. Is there a tutorial or reference example that shows:
    “Plan a trajectory with cuMotion and execute it on a real robot”?

We would like to understand the recommended architecture because our goal is to develop in Isaac Sim and deploy the same workflow on the real robot.

Thanks!

Additional observation:

From the NVIDIA documentation we understand that cuMotion is typically used as a planner within MoveIt, while trajectory execution is usually handled through FollowJointTrajectory / ros2_control.

On our AI Accelerator setup we can successfully generate trajectories with:

  • /IP192168000100/ur_cumotion/plan_motion

  • /IP192168000100/ur_cumotion/plan_multi_goal_motion

and trajectories are published on:

  • /IP192168000100/ur_cumotion/last_planned_trajectory

However, we cannot find a FollowJointTrajectory action server.

Instead we see:

  • logic_program_compiler

  • set_logic_program_state

  • logic_program_actual_state

  • program_running

  • program_state

This makes us wonder whether trajectory execution on the AI Accelerator is implemented through the PolyScope X Logic Program infrastructure.

Is that correct?

If so, is there a reference example showing the complete path from ur_cumotion planning to execution on a real robot?