How to test robot_runner when run(self, policy: BasePolicy) is not implemented?

[1024AILab]

Hi, I noticed that in the robot_runner module, the run(self, policy: BasePolicy) method is not implemented, which makes testing difficult. However, I see that in the DP3 source code, similar run functions are implemented for Metaworld, Dexart, and Adroit. Could you please clarify:

Is the run method in robot_runner intentionally left unimplemented for now?

If so, how should I properly test policies with robot_runner in the current state?

Are there any temporary workarounds or examples I can follow from the existing run implementations in Metaworld, Dexart, or Adroit?

Thanks for your guidance!

[1024AILab]

SADG obs have four input, but DP3 only have two input. How can I match the input for DP3 ? Thank you~

[wayrise]

Thanks for your issue.

As for the first issue, our baseline DP and DP3 refers to the implementation of the original codebase, which can be found in DP and DP3. So there are some codes (i.e. implementations in Metaworld, Dexart, or Adroit) belong to the original codebase, but not used in DexGarmentLab. Please refer the content in picture below for DP inference.

As for the second issue, the 'agent_pos' in SADP is just the 'agent_pos' in DP3, the 'environment_point_cloud' in SADP is just the 'point_cloud' in DP3.

Hope this answer can solve your issue.

[1024AILab]

I got it, thank you~

[1024AILab]

@wayrise Can you help me check if the test code I wrote for DP3 and DP is correct? I finished training DP3 and DP, but during testing, all the results from DP3 were wrong. DP couldn't be tested as it got stuck in an infinite loop, and there were no error messages in the printed output. Here's the inference video for DP3 and my code. Thank you so much!

https://github.com/user-attachments/assets/556617f8-4bec-4d16-9b1b-585d7e8f55e7
https://github.com/user-attachments/assets/2ae62f55-b745-44fb-9d9c-2806e0d94db3

Hang_Dress_DP.py
Hang_Dress_DP3.py