Approximate Convex Decomposition-based Whole-Body Trajectory Optimization for Robots in Dense Environments

Abstract

Whole-body planning is critical for enabling robots to navigate effectively in complex and dense environments. Traditional obstacle-based planning methods methods often restrict the representation of both robots and obstacles to simple convex polyhedra. This limitation may fail to adequately address intricate geometries of real-world obstacles involved in constructing compact convex polyhedral envelopes around more intricate obstacle shapes found in such environments. In this paper, we propose an approximate convex decomposition (ACD) based method to generate convex polyhedral maps that effectively represent the non-convex shapes of robots as assemblies of multiple convex objects. Furthermore, we propose a differentiable convex polyhedron collision evaluation method to facilitate collision detection. Extensive experiments demonstrate that our method not only enhances the accuracy of collision detection in cluttered environments but also expands the potential applications of robotics in complex scenarios.