SAC#

RoboVerse provides CleanRL-based SAC implementation for continuous control tasks.

Usage#

# SAC training with RoboVerse environment
python roboverse_learn/rl/clean_rl/sac.py --task reach_origin --robot franka --sim mjx --num_envs 128

Configuration#

Check the file header in roboverse_learn/rl/clean_rl/sac.py for available configuration options including:

  • Task selection (--task)

  • Robot type (--robot)

  • Simulator backend (--sim)

  • Training hyperparameters (--num_envs, --learning_rate, etc.)

Features#

  • Based on CleanRL SAC implementation

  • Vectorized environment support with RoboVerse

  • Automatic entropy tuning

  • Episode tracking without relying on info dicts