Probabilistic Pick and Place Planning instead of Pick then Place Planning

Mohanraj Devendran Shanthi, Tucker Hermans

Keywords: pick and place, manipulation, grasping, grasp inference, differentiable probabilistic inference

TL;DR: Formulating pick and place into clutter as a joint inference, and experimentally validating that joint inference is significantly better than sequential pick then place

Abstract: Robotic pick and place is a key problem of autonomous manipulation. The success of a pick and place execution is conditioned on the success of the picking, hence it is natural to think about the pick and place problem jointly. In this paper, we formulate the pick and place problem as maximizing the joint probability of pick and place success. Benchmarking the joint inference against sequential pick then place approach and random sampling show that joint inference over pick and place significantly improves the success rate over the baselines.

Submission Number: 3