Robot navigation algorithms using learned spatial graphs

Conference · Tue Jan 01 04:00:00 EST 1985

OSTI ID:6389782

Iyengar, S S; Jorgensen, C C; Rao, S V.N.; Weisbin, C R

Finding optimal paths for robot navigation in known terrain has been studied for some time but, in many important situations, a robot would be required to navigate in completely new or partially explored terrain. We propose a method of robot navigation which requires no pre-learned model, makes maximal use of available information, records and synthesizes information from multiple journeys, and contains concepts of learning that allow for continuous transition from local to global path optimality. The model of the terrain consists of a spatial graph and a Voronoi diagram. Using acquired sensor data, polygonal boundaries containing perceived obstacles shrink to approximate the actual obstacles' surfaces, free space for transit is correspondingly enlarged, and additional nodes and edges are recorded based on path intersections and stop points. Navigation planning is gradually accelerated with experience since improved global map information minimizes the need for further sensor data acquisition. Our method currently assumes obstacle locations are unchanging, navigation can be successfully conducted using two-dimensional projections, and sensor information is precise.

• Research Organization: Oak Ridge National Lab., TN (USA)
• DOE Contract Number: AC05-84OR21400
• OSTI ID: 6389782
• Report Number(s): CONF-851245-2; ON: DE86002238
• Country of Publication: United States
• Language: English

Related Subjects

42 ENGINEERING
420100* -- Engineering-- General Engineering-- (-1987)
ALGORITHMS
DATA ACQUISITION SYSTEMS
MATHEMATICAL LOGIC
NAVIGATION
PROBES
ROBOTS