Launch sequence for evaluation trial with aic_mujoco

Competitions AI for Industry Challenge
1

Hello, following the README in aic_mujoco, I am able to run the teleop in the MuJoCo scene just fine. Now, I want to try the end-to-end example such as WaveArm or RunACT. What should be the expected launch sequence to do it? I have a local build from source. I tried the following but it results into a folded initial position of the robot and flickering image data (in Rviz).

Terminal 1: ros2 run rmw_zenoh_cpp rmw_zenohd

Terminal 2: ros2 launch aic_bringup aic_gz_bringup.launch.py ground_truth:=false start_aic_engine:=true gazebo_gui:=false

Terminal 3: ros2 launch aic_mujoco aic_mujoco_bringup.launch.py

Terminal 4: ros2 run aic_model aic_model --ros-args -p use_sim_time:=true -p policy:=aic_example_policies.ros.WaveArm

I have verified that I am using the same RMW config (Zenoh + SharedMem) for all four sessions.

Let me know the correct instructions. Meanwhile, if I find it, I will share it here.

Finally, to the organizers, thank you for organizing such a well constructed engineering challenge!

2

Update:

For now, I have settled with the understanding that evaluations (spawning of the plugs and ports, trial orchestration etc) will be run exclusively on gazebo environments whereas other simulator (Mujoco, Isaac sim) adaptors are simply to enable development of policies in that ecosystem. So I tweaked aic_gz_bringup to spawn plugs and ports (in addition to just task board and cable), exported that and converted to Mujoco description and started my Mujoco+Gym RL workflow.

1 Like
3

Hi @shrikadam

Thanks for diving deep into the toolkit!

You’ve arrived at the right conclusion. We’ll use Gazebo for eval and provide MuJoCo or IsaacLab environments for training if needed.