A Reactive Robot System for Find and Fetch Tasks in an Outdoor Environment

Book One

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 10

Robot Navigation

This paper describes the results of using a reactive control software architecture for a mobile robot retrieval task in an outdoor environment. The software architecture draws from the ideas of universal plans and subsumption’s layered control, producing reaction plans that exploit low-level competences as operators. The retrieval task requires the robot to locate and navigate to a donor agent, receive an object from the donor, and return. The implementation employs the concept of navigation templates (NaTs) to construct and update an obstacle space from which navigation plans are developed and continually revised. Selective perception is employed among an infrared beacon detector which determines the bearing to the donor, a real-time stereo vision system which obtains the range, and ultrasonic sensors which monitor for obstacles en route. The perception routines achieve a robust, controlled switching among sensor modes as defined by the reaction plan of the robot. In demonstration runs in an outdoor parking lot, the robot located the donor object while avoiding obstacles and executed the retrieval task among a variety of moving and stationary objects, including moving cars, without stopping its traversal motion. The architecture was previously reported to be effective for simple navigation and pick and place tasks using ultrasonics. Thus the results reported herein indicate that the architecture will scale well to more complex tasks using a variety of sensors.