About actions in record_episodes and imitate_episode

MarkFzp / act-plus-plus

Imitation learning algorithms with Co-training for Mobile ALOHA: ACT, Diffusion Policy, VINN

https://mobile-aloha.github.io/

MIT License

3.07k stars 564 forks source link

About actions in record_episodes and imitate_episode #62

Closed Cryon233 closed 3 days ago

Cryon233 commented 2 weeks ago

Hi, I'm trying to apply the ACT algorithm to other robotic arms I have a question as to why the action dimension of the dataset produced by record_sim__episodes.py is 14, but the action dimension of imitate_episode.py via load_data is 16 I noticed that the EpisodicDataset class created by load_data and get_norm_stats() added dummy_base_action to the action(padding two zeros after the action) causing this phenomenon, and I wondered why Thanks for a job well done!

TomWangzs commented 1 week ago

Since this is a mobile bimanual robot, in addition to the 14-dimensional arm position parameters, two additional dimensions for the base position parameters are also required.

Cryon233 commented 3 days ago

由于这是一款移动式双手动机器人，因此除了 14 维手臂位置参数外，还需要两个额外的基本位置参数维度。

Thanks very much!