Multi-layer path planning control for the simulation of manipulation tasks: Involving semantics and topology

Abstract

The industrial and research communities show increasing interest in using automatic path planning techniques for the simulation of manipulation tasks. Automatic path planning, largely explored by the robotics community over the past 30 years, computes the trajectories of robots or manipulated parts. However, as techniques developed so far use mostly purely (and large) geometric models, they may fail, produce a trajectory of little relevance, or lead to very high computation times, when facing complex or very constrained environments. Involving higher abstraction level information should lead to better relevance of the simulation. In this paper, we propose a novel path planning technique relying on an original multi-layer environment model containing geometrical, topological and semantic layers. A first coarse planning step at the topological and semantic layers and a fine planning step at the local and semantically characterized geometrical layer form the path planning process. Experimental full-scale results show increased control on the planning process, leading to much lower computation times and increased relevance of the computed trajectory.