Inverse Kinematics in MuJoCo

Since, MuJoCo does not have motion planners I had to try some custom and available options for MuJoCo. Currently I have tested the following.

• [ ] Custom IK solutions based on Gradient Descent, but this currently gives errors in end-effector orientation. (Reference)
• [ ] On Google Deep Mind there was a inverse_kinematics function from the dm_control library but that seems to have some convergence issue. (Reference)
• [ ] The KDL, TrackIK infrastructure are more for the ROS architecture unable to integrate with MuJoCo getting errors which have no references online.

2024-11-30 12-21-53.mkv

For Current : I am generating the joint angles needed from the previous ROS-Gazebo architecture and hard coding the joint angles for the PD Controller.

The performance of the same is attached beside.