How do you compute the inverse kinematics?

[elenacliu]

First thank you for all your excellent work! When we want the end-effector to go somewhere in the world frame, we just call the goto_pose function, but where do you compute the ik?

[mohitsharma0690]

We don't directly compute IK (although you can add that functionality if you wanted and we would be more than happy to integrate it), instead we rely on Franka's IK (by directly using their position control). Alternatively, we implement an OSC controller that uses jacobians that you can use as well.

[elenacliu]

Thank you. So the position control is implemented by franka_ros or libfranka? I will try the OSC controller!

[elenacliu]

sorry to bother you, but I noticed that somoetimes the arm would reach joint limits when executing goto_pose or goto_joints in the trajectory, what can I do then?

p.s. It's not because the virtual wall because I have modified the value in franka-interface repo and pass virtual_walls=False in my code.

[mohitsharma0690]

well if the target_configuration ($q_{target}$) is close to the joint limit of the robot then there is not much you can do. But since we do linear interpolation between $q_{src}$ and $q_{target}$, there can be some edge cases where the robot reaches some limit during interpolation. For such cases, you can always break up the target pose into intermediate poses and manually combine them, this ensures that you're within the configuration space of the robot and not close to singularity.

For more complex scenarios, you may want to implement some traj opt based system which takes forward kinematics into account and can reason about these issues.

[elenacliu]

Thank you for your thorough explanation! By the way, can I use other planner and controller with frankapy, like MoveIt and joint trajectory controller provided by ROS?

upd: I see a PR in this repo (#15) which tries integrating moveit.