Accession Number:

ADA220168

Title:

New Heuristic Algorithms for Efficient Hierarchical Path Planning

Descriptive Note:

Research rept.

Corporate Author:

STANFORD UNIV CA DEPT OF COMPUTER SCIENCE

Personal Author(s):

Report Date:

1989-04-01

Pagination or Media Count:

44.0

Abstract:

One of the ultimate goals of robotics research is to create autonomous robots. Progress toward this goal requires advances in many domains, including automatic motion planning. The basic problem in motion planning is to construct a collision-free path for a moving object among fixed obstacles. Several approaches have been proposed, including cell decomposition, retraction, and potential field. Nevertheless, most existing planners still lack efficiency, or reliability, or both. In this paper, we consider one of the most popular approaches to path planning hierarchical approximate cell decomposition. We propose a set of new algorithms for constructing more efficient and reliable path planners based on this general approach. These algorithms concern the hierarchical decomposition of the robots configuration space into rectangloid cells, and the search of the connectivity graphs built at each level of decomposition. We have implemented these algorithms in a path planner and experimented with this planner on various examples. Some are described in the paper. These experiments show that our planner is significantly faster than previous planners based on the same general approach.

Subject Categories:

  • Cybernetics

Distribution Statement:

APPROVED FOR PUBLIC RELEASE