wty-yy/go2_rl_gym_data

Go2 RL Data

Checkpoint files for go2_rl_lab, go2_rl_gym and go2_rl_robotlab.

📦 Files

The table below includes released checkpoints and their RoboGauge evaluation results. Link points to the corresponding Hugging Face file or folder page. Last Updated records the artifact release date; legacy entries without a dated name use their initial Hugging Face release date. Scores are rounded to four decimal places from evaluation YAML files; directory names retain their original score labels.

Model	Best Train Step	Score	Tracking	Safety	Quality	Level	Link	Last Updated
go2_moe_cts (RobotLab v4.2)	176k	0.6984	0.7055	0.7850	0.7845	8.62	go2_moe_cts_v4.2_176k_0.6984	2026-06-26
go2_mixed_moe_cts (go2_rl_lab mixed1.3.2resume)	82.5k	0.6930	0.7004	0.7818	0.7818	8.40	mixed1.3.2resume_82.5k_0.6929	2026-10-17
go2_mixed_moe_cts (go2_rl_lab v1.3)	89k	0.6903	0.6950	0.7773	0.7775	8.34	v1.3_8192_89k_0.69	2026-10-17
go2_mixed_moe_cts (go2_rl_lab mixed1.3.2resume)	74k	0.6866	0.6944	0.7763	0.7762	8.34	mixed1.3.2resume_74k_0.687	2026-10-17
go2_moe_cts (RobotLab v3.3)	185k	0.6828	0.6785	0.7552	0.7645	8.17	go2_moe_cts_v3.3_185k_0.6828	2026-06-16
go2_moe_cts (go2_rl_gym)	137k	0.6713	0.6669	0.7857	0.7392	7.85	go2_moe_cts_137000_0.6713	2026-01-31
go2_ac_moe_cts	115k	0.6509	0.6442	0.7644	0.7149	7.52	go2_ac_moe_cts_115k_0.6509.pt	2026-01-31
go2_mcp_cts	91k	0.6399	0.6355	0.7542	0.7058	7.41	go2_mcp_cts_91k_0.6399	2026-01-31
go2_moe_ng_cts	79k	0.6519	0.6447	0.7639	0.7186	7.56	go2_moe_ng_cts_79k_0.6519	2026-01-31
CTS vanilla	103.5k	0.5786	0.5755	0.7066	0.6624	6.83	go2_cts_vanilla2_103.5k_0.5786	2026-01-31
HIM	21k	0.5379	0.5453	0.6476	0.6050	6.19	go2_him_21k_0.5379.pt	2026-01-31
DreamWaQ	119.5k	0.5054	0.5105	0.6149	0.5730	5.74	go2_dwaq_119.5k_0.5054.pt	2026-01-31

Additional self-collision-enabled result:

Model	Best Train Step	Score	Download	Last Updated
go2_moe_cts_high_slope_thre	164k	0.6715	go2_moe_cts_high_slope_thre_164k_0.6715_20260419	2026-04-19

🚀 Usage

File types:

• policy.pt: TorchScript policy for Sim2Sim deployment.
• policy.onnx: ONNX policy for C++ / Sim2Real deployment.
• policy.pkl: raw policy weights.
• model_*.pt: full training checkpoint.
• exported/config.yaml: deployment configuration for go2_rl_lab policies.
• params/: saved training configuration for go2_rl_lab checkpoints.
• robogauge_results_*.yaml: RoboGauge evaluation results for go2_rl_lab checkpoints.
• single *.pt files at repository root: standalone checkpoint files kept as released.

Example paths:

# Sim2Sim policy
./go2_moe_cts_137000_0.6713/policy.pt

# ONNX deployment policy
./go2_moe_cts_137000_0.6713/policy.onnx

# Full training checkpoint
./go2_moe_cts_self_103.5k_0.6669/model_103500.pt

# go2_rl_lab ONNX deployment policy (config.yaml is in the same directory)
./go2_rl_lab/2026-10-13_15-35-09_mixed1.3.2resume_82.5k_0.6929/exported/policy.onnx

# go2_rl_lab full training checkpoint
./go2_rl_lab/2026-10-13_15-35-09_mixed1.3.2resume_82.5k_0.6929/model_82500.pt

For training, evaluation, export, and deployment commands, see the main repository go2_rl_gym README.

🔎 Details

• Most directory entries contain exported policies.
• Each go2_rl_lab/<run>/ directory contains a full training checkpoint, TorchScript and ONNX policies under exported/, deployment and training configurations, and RoboGauge evaluation results.
• go2_moe_cts_high_slope_thre_164k_0.6715_20260419 contains both exported policies and the full model_164000.pt.
• go2_moe_cts_self_103.5k_0.6669 contains exported policies and model_103500.pt.
• On the Hugging Face web page, folder entries should use tree/main/... URLs for navigation.

📄 Citation

If you find our work helpful, please cite:

@INPROCEEDINGS{WuT-RSS-26,
    AUTHOR    = {Tianyang Wu AND Hanwei Guo AND Yuhang Wang AND Junshu Yang AND Xinyang Sui AND Jiayi Xie AND Xingyu Chen AND Zeyang Liu AND Xuguang Lan},
    TITLE     = {{Toward Reliable Sim-to-Real Predictability for MoE-based Robust Quadrupedal Locomotion}},
    BOOKTITLE = {Proceedings of Robotics: Science and Systems},
    YEAR      = {2026},
    ADDRESS   = {Sydney, Australia},
    MONTH     = {July},
    DOI       = {10.15607/RSS.2026.XXII.156}
}