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

Sipu Ruan, Weixiao Liu, Xiaoli Wang, Xin Meng, Gregory S Chirikjian

Keywords: Probability and Statistical Methods, Learning from Demonstration, Motion and Path Planning, Service Robots

TL;DR: A novel probabilistic method for learning demonstrated trajectories in the 6D workspace includes orientation and position.

Abstract: This paper proposes a novel framework for learning skills from one or a few demonstrations and generalizing skills to unseen objects, environments, and robots. The proposed learning-from-demonstration method, named “PRobabilistically- Informed Motion Primitives (PRIMP)”, captures motion features by learning the probability distribution of the end effector trajectories in the 6D workspace that includes both positions and orientations. The learned trajectory distribution can adapt to new situations, such as novel via points and changes in the viewing frame. The method itself is robot-agnostic, in that the learned distribution can be transferred to another robot with the adaptation to its workspace density. The learned trajectory distribution is then used to guide an optimization-based motion planning algorithm to further help the robot avoid novel obstacles that are unseen during demonstrations while keeping the learned motion features. The proposed method is evaluated by several sets of benchmark experiments. PRIMP runs more than 5 times faster than existing state-of-the-art methods while generalizing trajectories more than twice as close to both the demonstrations and novel via points. It is then combined with our lab’s robot imagination method that learns object affordances, illustrating the applicability of PRIMP to learn tool use through physical experiments.

Submission Number: 8