Probabilistic constraints for optimization-based motion planning

Abstract

We present a novel method to formulate constraints over probability distributions within optimization-based motion planning, enabling robots to plan for uncertainty in dynamic environments. Typical approaches for motion planning in dynamic environments involve predictions of the environment (e.g., moving obstacles), so that the robot can plan around anticipated environmental changes. Uncertainty in predictions is a challenge that pervades these approaches. As such, properly modeling environmental uncertainty is crucial to obtain valid motion plans. We propose a general method of propagating uncertainty in environmental predictions to a probabilistic constraint in optimization-based motion planning. We demonstrate the efficacy of this approach for uncertainty in predicted human motions in the context of simultaneous manipulation in a shared workspace with a 7-DoF robot arm.