Any-shape Real-time Replanning via Swept Volume SDF

Abstract

Existing robotic trajectory planning frameworks typically approximate the robot’s geometry and environmental constraints. While this improves computational efficiency, it sacrifices the solution space and frequently encounters failure in confined environments. However, attaining a precise geometric representation and a continuous collision-free trajectory usually necessitates greater computational expenditure. This paper proposes a methodology that utilizes the concept of swept volume to address the identified limitations. The implementation of an efficient Swept Volume Signed Distance Field computation algorithm and a B-spline trajectory representation results in a significant increase in computational efficiency while maintaining strict safety guarantees. The proposed method combines the advantages of efficiency and maximal exploitation of the solution space. Additionally, it ensures continuous obstacle avoidance, achieving real-time 10Hz replanning performance on i50000 NUC11TNK for arbitrarily shaped rigid objects in complex, unstructured environments.