Development Utilities

Use these utilities to update assets or clean the cache.

Reward-function reference plots

Generate PDF references for implemented reward terms with CPU Torch:

1. From a source checkout, install the plots extra (Matplotlib, NumPy, Torch, Seaborn).

2. Run this command to write the PDFs to outputs/reward_terms/:

   MPLBACKEND=Agg uv run --extra plots python - <<'PY'
   from pathlib import Path
   
   import matplotlib.pyplot as plt
   
   from srb.core.reward.scripts.plot_reward_terms import plot_library_final_order
   
   output = Path("outputs/reward_terms")
   output.mkdir(parents=True, exist_ok=True)
   plot_library_final_order(weight=1.0, limit=4.0, output_path=output / "reward_terms.pdf")
   plot_library_final_order(weight=-2.0, limit=64.0, output_path=output / "penalty_terms.pdf")
   plt.close("all")
   PY
   

Agg needs no display. PDFs show reward functions, not task dynamics or learned behavior. Running srb/core/reward/scripts/plot_reward_terms.py directly writes PDFs beside the script; the function above writes to outputs/.

For the manual TF trajectory visualizer, see the ROS integration guide.