Safe and Efficient Motion Planning in Robotic Manipulation through Formal Methods

Abstract

Manipulating rigid body objects in crowded environments poses significant challenges due to the need for rapid, real-time planning and the assurance of safe operational paths. The challenges come from varying shapes of the manipulated objects and high-dimensional nature of manipulators.

This thesis addresses these issues by developing (1) a mixed-integer linear programming (MILP)-based approach to plan safe paths for rigid-body objects; and (2) a learned control barrier function (CBF) tailored for manipulators with multiple degrees of freedom (DoF) and an associated framework CBF-RRT to enable efficient planning for robotic manipulators. Comprehensive experimental results have shown that the proposed methods outperform baseline methods, providing tools for improving the safety and efficiency of robotic manipulators in complex environments.