lerobot

/

xvla-widowx

like 4

Follow

LeRobot 3.65k

Thanks for sharing this checkpoint! I am trying to use it in a Mujoco simulation and wanted to confirm that:

1. the expected observation state is the EE6D in WidowX arm base frame, and
2. the output actions would be in the same frame, and absolute (not deltas)

Based on some testing, it appears that the actions returned are absolute (drawing dots at the xyz here), but I am having a lot of trouble confirming.

I used these camera views referring to the BridgeData-v2 description https://rail-berkeley.github.io/bridgedata/#:~:text=System%20Setup,saved%20at%20a%20640x480%20resolution. suggesting that the primary view is "overtheshoulder"