Safe and Efficient Motion Planning through Chance-Constrained Nonlinear Optimization

Abstract

Uncertainty is the harsh reality for robots deployed in the real world. Outside of a carefully-structured laboratory environment, neither the locations of obstacles nor the true state of the robot can be known with perfect certainty. This makes planning safe maneuvers challenging, particularly for robots with many degrees of freedom and rich geometry. Existing uncertainty-aware planners fall short by considering only uncertainty in the environment or uncertainty in the robots' state. In this thesis, we develop a chance-constrained trajectory optimization framework to address this gap in the state of the art, which we call Sequential Convex Optimization with Risk Allocation (SCORA). This planner is capable of solving challenging, high-dimensional motion planning problems while managing the risk due to uncertainty in the environment and in the robots own state. In addition, SCORA supports robots with nonlinear dynamics and arbitrary geometry, and it outperforms state-of-the-art planners in terms of both safety and planning time on a range of robotics tasks, including autonomous parallel parking, control of a mobile robot arm, and planning for multi-agent manipulation tasks.