Inverse kinematic for Mujoco
a-z-e-r-i-l-a opened this issue · 3 comments
Hi, I wanted to ask if your implementations also allow inverse kinematic for mujoco.
I am interested in getting joints' values (MjData.qpos) for a given the desired end-effector pose, and not necessarily wish to move the arm to that position (similar to PyBullet's inverse kinematic api).
Hello!
One of the standard ways is to make a mocap object and weld it to the part you want to control, I've added some files that do this, for reference.
It doesn't sound like this will address your use case, though. We do have an inverse kinematics planner, but it requires access to the transformation matrices of the robot, which is a feature not available currently in our Mujoco robot_config. I remember it not being straight-forward to find from the Mujoco API, but I'll look into it again as this would be a very nice feature to offer.
Hello! Had the chance to look at this today and was able to get it working. It should hopefully be merged into master soon, but if you check out the m4 branch there is now a docs/examples/Mujoco/path_planning_inverse_kinematics.py file that you can run that I believe is an example of how to address your use case. If you don't need the whole path you can just access the last configuration in the path with path_planner.position_path[-1] once you've called the generate function to create the path.
Hopefully this addressed your issues, I'll close this for now and we can re-open if it's not working for you.