Results on multi-task robot manipulation benckmarks

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

489 stars 49 forks source link

Results on multi-task robot manipulation benckmarks #37

Closed hnuzhy closed 4 months ago

hnuzhy commented 5 months ago

Hi, thanks a lot for releasing your awesome work. Recently, I'm doing experiments to tackle the multi-task robot manipulation problem. Two widely used benckmarks are RLBench and CALVIN. I'm wondering that if the proposed DP3 algorithm can be used for tackling this kind of problem or not? Or maybe it can be integrated with some high-level agent planner (e.g., PerAct) for decomposing long-horizon instructions to form a possible solver? Looking forward to your reply!

YanjieZe commented 4 months ago

Hi, thank you for your interest. It is surely possible to use DP3 in multi-task benchmarks. In our work, we do not put our focus on this, but it is definitely worth doing. Looking forward your results on this.