mees / calvin

CALVIN - A benchmark for Language-Conditioned Policy Learning for Long-Horizon Robot Manipulation Tasks
http://calvin.cs.uni-freiburg.de
MIT License
366 stars 55 forks source link

New policy to the benchmark #83

Open StarCycle opened 4 months ago

StarCycle commented 4 months ago

Hi guys😊,

Can I add my new policy to CALVIN benchmark? On ABC->D, my policy is current SOTA (average length ~3.55). I have not tested ABCD->D yet but I can do so if you want.

The code, data, checkpoints and logs are fully open-source in https://github.com/EDiRobotics/GR1-Training. It's modified on GR-1 implementation. You can easily evaluate it following the commands in the Github page.

Please let me know if my setting is wrong😣

Best,

Zhuoheng