Run an Environment¶
This guide runs a registered AM-Bench environment independently of a learned policy. Use it to inspect scene startup, observations, actions, cameras, and simulator performance before adding data collection or evaluation.
Complete Verify Installation, then choose an exact ID with Find an Environment. Run the commands below from the repository root with env_isaaclab active.
Run a bounded headless session¶
Check GPU availability before starting Isaac Sim:
Run one environment for a bounded interval:
timeout --signal=INT 30s python scripts/environments/zero_agent.py \
--task PressButton-Am-EE-Abs-PID-Direct-v0 \
--num_envs 1 \
--headless \
--device cuda:0
Replace --task only with an ID printed by scripts/environments/list_envs.py. The script prints the observation and action spaces, then steps continuously with a neutral action until interrupted. Task completion is not expected.
Open a simulator window¶
Omit --headless when the machine has a working graphical session:
python scripts/environments/zero_agent.py \
--task PressButton-Am-EE-Abs-PID-Direct-v0 \
--num_envs 1 \
--device cuda:0
Close the Isaac Sim window or interrupt the process to stop the run. A remote machine requires a supported display or streaming setup; a successful headless run does not prove that GUI rendering is configured.
Record a camera check¶
Enable video only after the basic headless session works:
timeout --signal=INT 30s python scripts/environments/zero_agent.py \
--task PressButton-Am-EE-Abs-PID-Direct-v0 \
--num_envs 1 \
--video \
--camera_names ee_camera \
--headless \
--device cuda:0
The script prints VIDEO_PATH after a clean shutdown. Video confirms that the named camera and encoder work; it does not validate task success. Camera availability depends on the selected robot profile.
Scale only after one environment works¶
Increase --num_envs only after the one-environment run is stable. More environments and cameras increase GPU memory use. On a single-GPU machine, keep one heavy Isaac Sim process active at a time and confirm that the previous process has exited before starting another.
Continue with Collect Demonstrations to record data, Evaluate a Supported Policy, or Troubleshooting when startup or stepping fails.