I'm trying to use mocap to control the franka arm like fetch_pick_and_place.env.
After I created this environment and tested it, I noticed that the arm drifted abnormally each time I initiated this env, even though I gave the initial configuration for data.qpos and the initial pose for mocap.
Currently, I think the triggering reason might be that the given initial configuration is different from that solved by the built-in inverse kinematics algorithm. For comparison, I also initialized fetch_pick_and_place.env in the same way and did not notice a similar swing. If the key factor is the inverse kinematics algorithm, why can the fetch arm maintain almost the same configuration at each initial stage? The fetch arm is also redundant.
Here is the video to describe the loop to initialize fetch_pick_and_place.env
Hi!
I'm trying to use mocap to control the franka arm like
fetch_pick_and_place.env.After I created this environment and tested it, I noticed that the arm drifted abnormally each time I initiated this env, even though I gave the initial configuration for
data.qposand the initial pose formocap.Here is a video to illustrate my dilemma
https://github.com/Farama-Foundation/Gymnasium-Robotics/assets/72594017/60121268-ea59-454f-b3c5-242943ee3e75
in which I repeat the startup of this env.
Currently, I think the triggering reason might be that the given initial configuration is different from that solved by the built-in inverse kinematics algorithm. For comparison, I also initialized
fetch_pick_and_place.envin the same way and did not notice a similar swing. If the key factor is the inverse kinematics algorithm, why can the fetch arm maintain almost the same configuration at each initial stage? The fetch arm is also redundant.Here is the video to describe the loop to initialize
fetch_pick_and_place.envhttps://github.com/Farama-Foundation/Gymnasium-Robotics/assets/72594017/c17d7e15-b112-40bd-90ea-6ab5eb49f00f
Also, maybe I missed some other factors.
Could you give me some advice on how to solve it?
Thanks in advance!