--- name: rl-isaaclab description: "End-to-end IsaacLab RL workflow in moleworks_ext: run with isaaclab.sh -p, local smoke tests, cluster submit/monitor/debug on Euler, sync results, and play policies." --- # RL + IsaacLab Workflow (moleworks_ext) Single-skill workflow for IsaacLab-based RL: local smoke test → cluster submit → monitor/debug → sync → playback. ## 1) Always Run IsaacLab Scripts via Wrapper **IMPORTANT**: Any script that touches Isaac Sim/Lab must use: ```bash /workspace/isaaclab/isaaclab.sh -p [args] ``` Use this for: - Isaac Sim/Lab imports - USD/PhysX - GPU/RTX - training/testing scripts ### pxr Import Errors If you hit `ImportError: No module named 'pxr'`, load the app launcher **inside the file**: ```python from isaacsim import SimulationApp simulation_app = SimulationApp({"headless": True}) from pxr import Usd, UsdGeom ``` ## 2) Local Smoke Test (ALWAYS first) Disable W&B for local smoke tests to avoid junk runs. ```bash export WANDB_MODE=disabled /workspace/isaaclab/isaaclab.sh -p scripts/rsl_rl/train.py \ --task --num_envs 4 --max_iterations 3 --headless unset WANDB_MODE ``` Example (excavation3d_w_cabin): ```bash export WANDB_MODE=disabled /workspace/isaaclab/isaaclab.sh -p scripts/rsl_rl/train.py \ --task Moleworks-Isaac-m445-digging-3D-w-cabin --num_envs 4 --max_iterations 3 --headless unset WANDB_MODE ``` ## 3) Submit to Euler Cluster ```bash JOB_TIME=30m NUM_GPUS=2 GPU_TYPE=rtx_3090 ./docker/cluster/cluster_interface.sh job \ --task --num_envs 64000 --max_iterations 10000 ``` Example (excavation3d_w_cabin, 24h, 1 GPU): ```bash JOB_TIME=24h NUM_GPUS=1 GPU_TYPE=rtx_3090 ./docker/cluster/cluster_interface.sh job \ --task Moleworks-Isaac-m445-digging-3D-w-cabin --num_envs 64000 --max_iterations 10000 ``` Custom entrypoint (optional): ```bash CLUSTER_EXECUTABLE=scripts/mole_environments/grading/train_grading.py \ JOB_TIME=2h NUM_GPUS=2 GPU_TYPE=rtx_3090 ./docker/cluster/cluster_interface.sh job \ --task Moleworks-Isaac-m445-grading --num_envs 64000 --max_iterations 10000 ``` ## 4) Monitor Jobs (Euler) ```bash ssh euler 'squeue -u $USER' ``` Job states: - RUNNING: executing - PENDING: waiting on resources - FAILED/NODE_FAIL: crashed ## 5) Debug Failed Jobs ```bash ssh euler 'find /cluster/scratch/$USER -name "slurm-*.out" -mmin -60' ssh euler 'tail -100 /cluster/scratch//moleworks_ext_/slurm-.out' ssh euler 'cat /cluster/scratch//moleworks_ext_/slurm-.err' ssh euler 'tail -50 /cluster/scratch//moleworks_ext_/slurm-.err' ssh euler 'grep -n "bind mount detected" /cluster/scratch//moleworks_ext_/slurm-.out | head' ``` **Note**: `.err` is where Python tracebacks are. ## 6) Sync Results ```bash ./docker/cluster/sync_experiments.sh # or to save space ./docker/cluster/sync_experiments.sh --remove ``` ## 7) Play Policy Locally ```bash /workspace/isaaclab/isaaclab.sh -p scripts/rsl_rl/play.py \ --task \ --checkpoint logs/rsl_rl///model_150.pt \ --num_envs 1 ``` ## RL Debugging Tips - Init order matters: buffers must exist before ObservationManager uses them. - CurriculumManager uses `compute()` (EventManager uses `apply()`). - Keep rewards bounded: `exp(-penalty)` keeps in (0,1]. - Reference `excavation3d_w_cabin` for established API patterns. - If behavior is odd: check reward farming, velocity distribution, and speed penalties. ## tmux Requirement When running IsaacLab scripts or ROS commands, open a **new tmux window** so output persists: ```bash tmux new-window -n rl-isaaclab /workspace/isaaclab/isaaclab.sh -p scripts/rsl_rl/train.py --task --num_envs 4 tmux capture-pane -p ```

rl-isaaclab

Scan your own AI components