[Feature] adpat agibot a2 dof12 into rl locomotion

[wangbeiabcd]

Description

Adapting the 12-DOF version of the agibot a2 robot to the rl framework in RoboVerse learn, supporting locomotion training using the IsaacGym simulator and sim2sim using the MuJoCo simulator.

Currently, only the basic training and sim2sim pipeline have been adapted. Reward parameters still need some adjustments to achieve better walking effects. When the reward parameters are properly adjusted, documentation and more content will be supplemented.

Fixes:

1. Fixed the path for model training and evaluation
2. Fixed pop() logic in the ppo.py
3. Fixed IsaacGym null check
4. Fixed device settings and model weights loading in the eval.py

Type of change

Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds functionality)

How to test

Train in isaacgym from scratch,

python roboverse_learn/rl/rsl_rl/ppo.py --task walk_agibot_a2_dof12 --sim isaacgym --num-envs 4096 --robot agibot_a2_dof12 --headless

Eval in isaacgym from a checkpoint,

python roboverse_learn/rl/rsl_rl/eval.py --task walk_agibot_a2_dof12 --sim isaacgym --num-envs 1 --robot agibot_a2_dof12 --resume 20251225_171031 --checkpoint -1

If you want to use mujoco to do sim2sim, just change isaacgym to mujoco and reduce num_envs to 1.

Screenshots / Videos

2025.12.26.14.53.23.webm

Checklist

• I have run the pre-commit checks with pre-commit run --color=always --all-files
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have added my name to the CONTRIBUTORS.md or my name already exists there

[github-actions]

✅ Privileged CI passed on branch pr-736.

Test Suite Results:

• ✅ General tests
• ✅ MuJoCo tests
• ✅ Sapien3 tests
• ✅ IsaacSim tests
• ✅ IsaacGym tests

📦 Download test logs

🔗 View full details