A Programmable Near-Infrared Proximity Detector For Robot Navigation

Abstract

One of the most significant issues hindering the development of autonomous mobile robotic systems today is the lack of appropriate sensors for collecting high-resolution geometric information describing the robot's surroundings. Scanning laser rangefinders have found use in this regard, primarily in outdoor scenarios, and numerous configurations are being developed by several companies. The physical size, power consumption, and high initial cost of these prototype units make them somewhat impractical for near-term utilization on the smaller platforms typical of indoor applications. Simpler and much less expensive ultrasonic ranging systems have been employed with some limited success, principally for purposes of collision avoidance. Such systems suffer markedly, however, from numerous problems associated with the temperature dependence of the speed of sound in air, extremely slow wave propagation velocities, specular reflection of the emitted wavefront, and especially poor angular resolution due to beam divergence. This article discusses the design and testing of a low-cost, programmable near-infrared proximity detector intended specifically to address some of those problems when used for navigational purposes in a complementary fashion with acoustical ranging systems. The experimental prototype was installed on a mobile robot, and both sonar and near-infrared data were captured on various scans as the robot was moved through a room full of obstacles. By combining the best features from each sensor, a representation can be built that is more accurate than if either sensor were used alone. A second representation, based on the curvature primal sketch, can then be obtained from this refined map. This allows for matching between views, and also for creating an interface to a higher level program, such as a path planner.