search

YanjieZe / 3D-Diffusion-Policy

[RSS 2024] 3D Diffusion Policy: Generalizable Visuomotor Policy Learning via Simple 3D Representations
https://3d-diffusion-policy.github.io
MIT License
402 stars 37 forks source link

Question about run in simulate env #30

Closed xbkaishui closed 4 months ago

xbkaishui commented 4 months ago

Hi, thanks for your great work, I have a question, How to simulate training in a simulation environment? I can't find more details in code or document, is Imitation learning does not necessarily require a simulation environment?

YanjieZe commented 4 months ago

Hi, simulation environments are not used for training, but only for evaluation. Imitation learning only trains a policy, but the imitation learning loss does not actually reflect how well the policy is. Therefore, simulation environments are needed.

I think this is very common practise in robotics, and is not we propose for the first time.

xbkaishui commented 4 months ago

Thank you for your response. I am a beginner in the field of robotics and would like to ask how to set up a simulation environment. Do you have any recommended articles?

YanjieZe commented 4 months ago

You could start by running some standard sim envs. For example, ManiSkill: https://maniskill2.github.io/

xbkaishui commented 4 months ago

ok thanks

zackzhao commented 4 months ago

Hi, simulation environments are not used for training, but only for evaluation. Imitation learning only trains a policy, but the imitation learning loss does not actually reflect how well the policy is. Therefore, simulation environments are needed.

I think this is very common practise in robotics, and is not we propose for the first time.

Hi, I also have a same question. Can we use human demonstration in sim envs? If yes , and can you give some suggestion how to use sim envs to collect data?