Isn’t the problem of robotic kinematics very much unsolved? I seem to remember hearing that kinematics are a major barrier in the path towards effective robots. Are there promising ml approaches to the problem?