Graduated Fidelity Motion Planning

This paper presents an approach to differentially constrained robot motion planning and efficient replanning. Satisfaction of differential constraints is guaranteed by the search space which consists of motions that satisfy the constraints by construction. Any systematic replanning algorithm, e.g. D*, can be utilized to search the state lattice to find a motion plan that satisfies the differential constraints, and to repair it efficiently in the event of a change in the environment. Further efficiency is obtained by varying the fidelity of representation of the planning problem. High fidelity is utilized where it matters most, while it is lowered in the areas that do not affect the quality of the plan significantly. The paper presents a method of modifying the fidelity between replans, thereby enabling dynamic flexibility of the search space, while maintaining its compatibility with replanning algorithms. The approach is especially suited for mobile robotics applications in unknown challenging environments. We successfully applied the motion planner on a real robot: the planner featured 10Hz replan rate on minimal computing hardware [2], while satisfying the car-like differential constraints. In recent years there has been a growing interest in efficient motion replanning. Real mobile robot applications face challenges due to scarce and uncertain perception information. In order to facilitate planning a robot’s motion given such challenges, dynamic replanning algorithms were developed [3]. Such algorithms incorporate updated perception information and modify the motion plan accordingly, while reusing previous computation. This work introduces efficient replanning to motion planning under differential constraints that is based on searching a state lattice, a directed cyclic graph that encodes the constraints by construction [2]. Substantial computation is performed off-line to determine the connectivity of edges that represents the differential constraints. This allows fast planning (on-line) by utilizing standard search algorithms in this graph, while naturally satisfying the constraints. In order to satisfy the differential constraints, relatively high dimensionality of the state lattice may be required. Deterministic search in this setting can be computationally Copyright c © 2011, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. costly. This cost is especially significant in outdoor robotics applications, as they pose complicated planning problems, in particular due to complex environments. This paper addresses this limitation by managing the complexity of the search through modification of the fidelity of representation. The search space consists of one or more arbitrary regions of different fidelities. Lower fidelity of representation results in faster search, but higher fidelity results in better quality solutions. The approach is closely related to multi-resolution planning [1], but we use the term graduated fidelity to emphasize that the quality of representation is expressed not only as the resolution of state discretization, but also – more importantly in this setting – as the connectivity of edges between the vertices in the state lattice. Each region of the search space can be assigned a fidelity arbitrarily, yet practically this choice is guided by the region’s relevance for the planning problem and the availability of the environment information. In particular, it is often beneficial to utilize a high fidelity of representation in the immediate vicinity of the moving robot. Our method meets that need by allowing the regions of different fidelity to move or change shape arbitrarily. The contribution of this work is an improved state lattice search space that consists of regions of different fidelities of representation and allows the regions to move or change shape between replans. This search space remains compatible with standard search algorithms and is capable of producing motion plans that satisfy differential constraints without any post-processing. The state lattice structure is key to enable the reuse of computation which renders the presented algorithm efficient even on modest computing hardware. Graduated Fidelity Planning We extend the typical definition of the search graph by assuming that it consists of subgraphs G1,G2, . . . ,Gn. The arrangement of vertices and edges in each subgraph is assumed to be regular, but this arrangement may be different among subgraphs to reflect the differences in the fidelity of representation. This composite search space is maintained to remain a directed cyclic graph, so that replanning algorithms can be utilized to reuse previous computation while replanning. We define modifying a subgraph as arbitrarily changing 205 Proceedings, The Fourth International Symposium on Combinatorial Search (SoCS-2011)