Pi0 inference server success rate always 0 during simulation evaluation in liebro

[zhian17]

Hi Dexbotic team,

First of all, I want to express my appreciation for this project — it’s incredibly well-structured and the tools you provide make research and experimentation very convenient. Thank you for your hard work!

I encountered an issue while running the Pi0 inference server and performing simulation evaluation.

Steps to reproduce:

Start the Pi0 inference server:

CUDA_VISIBLE_DEVICES=0 python playground/benchmarks/libero/libero_pi0.py --task inference

Run the simulation evaluation:

docker run --rm -it
--gpus all
--network host
--privileged
-e NVIDIA_DRIVER_CAPABILITIES=all
-e MUJOCO_GL=egl
-v /home/xjzz/dexbotic/dexbotic-benchmark:/workspace
-v /home/xjzz/dexbotic/dexbotic-benchmark/results:/app/results
dexmal/dexbotic_benchmark
bash /workspace/scripts/env_sh/libero.sh /workspace/evaluation/configs/libero/example_pi0_libero.yaml

Problem:

During the evaluation, the success rate is always 0, even though the server starts without errors.

I would greatly appreciate it if you could provide guidance on how to resolve this issue. Thank you in advance for your help, and I hope to hear back soon!

[wwwzxxzw]

可以将模型checkpoints放在网盘上供我们下载吗，hf上不好下，谢谢

[zed0630]

Thank you for using our work.

To further pinpoint the issue, please add --verbose to the last command of scripts/env_sh/libero.sh to output more logs.

Once done, please provide the model inference output logs, the test service logs, and the result files. For reference: #8 (comment)