PRIMP: PRobabilistically-Informed Motion Primitives for Efficient Affordance Learning from Demonstration