Flexible autonomous robotic task scheduling using advanced RISC machines

Abstract

This research presents two methods for a group of Garcia robots to collaboratively decide which task to attend to, and to move to their respective locations. One method allows for a needed flexibility and dynamic analysis in this distributed system of any number of robots coupled with any number of target locations, but is tied too closely to distance measurements. The other method is the implementation of Peter Molnar’s approach, which is free from any specificity for determining preferences, but is shown to have some other limitations. The packet loss problem inherent of broadcast communications is addressed, as well. The robots make decisions interdependently with the other robots after the initial setup from a host computer of common environment variables such as a map, robot and target locations. A distributed network was established first for the robots so that information could be shared. Beacon messages were broadcasted at random intervals, while the robots handled TCP and UDP messages in separate threads. Once the needed data was acquired, each robot began their task scheduling decision-making process. If additional information was required from another robot still, a robot requested it. Our algorithm for task scheduling converged to an agreement rapidly, and resolved any possible gridlocks that occurred when two robots are exactly the same distance and with the highest preference for a single target. This method is found to be too closely tied into distance measures being used for preferences. Changes in the environment, such as new robots or targets, are handled during subsequent calls to decide for both methods. The result of this thesis shows two flexible and quick methods for task scheduling and assignment.