Dynamic Scheduling for Obstacle Avoidance in Mobile Robots

Abstract

In robotics applications the speed at which a task may be completed is often limited by computation rather than actuator or sensor throughput. This research concerns the efficient threads-based implementation of motor schema-based reactive navigation. One result of this implementation is the insight that best effort scheduling offers significant performance advantages on single- and multi-processor architectures over static priority or round-robin scheduling. However, robot safety requires guaranteed rather than best effort scheduling, and low latency is important, in addition, to allow the robot to respond quickly to the real world. This paper presents the design and initial implementation results of a distributed, real-time, multiprocessor, dynamic scheduler offering both good quality scheduling decisions as well as low scheduling latency for individual tasks. Lastly, based on our results, we present a design for schema-based robot control when higher-level planning tasks are integrated with lower level reactive navigation.