google-deepmind / google-deepmind/mujoco_playground
Rendering Environments During Training (GUI and WandB Integration)
- Dominant language
- Python
- Stars
- 2.2k
- Forks
- 359
- Avg merge
- 7d 3h
- Merged PRs (30d)
- 1
Description
Hello! I'm currently porting a custom environment (originally developed in the IsaacGymEnvs framework) to mujoco_playground. To facilitate debugging and better monitor training progress visually, I'd like to enable rendering during training sessions, and I'm wondering if these have already been implemented.
Specifically, I'm looking for:
1. **Real-time GUI Rendering:**
- Is there a straightforward method to launch training scripts (e.g., `train_jax_ppo.py`) with the full MuJoCo GUI active (non-headless), so I can visually inspect the environment as training occurs? This would help me debug for unintentional problems in the environment (e.g. check that my custom reset code is working)
2. **Periodic Rendering and Logging:**
- Are there existing utilities or recommended approaches for periodically rendering the environment during training and logging these visuals to Weights & Biases? This would greatly help track progress beyond simple reward metrics, allowing visual verification of environment resets, robot behaviors, and interactions.
Any guidance or pointers toward existing implementations or recommended best practices would be greatly appreciated. Thanks!
Contributor guide
Assessment
This issue has not been assessed yet.