Rapid Generation of Motion Plans for Modular Robotic Systems

Robots are needed to perform important tasks in field environments. A limitation to the practical use of robot systems is their high cost and long development times. It would be desirable to rapidly design and fabricate the hardware and software for these systems, on a time scale of days or weeks instead of years. An integral part of this system is a rapid generator of the motion plans that the robot will execute in order to perform the task. The goal of this research is to develop a methodology for generating these motion plans rapidly, and with as little user intervention as possible. This thesis describes one approach to rapid generation of motion plans for robotic systems operating in field environments. The approach assembles software modules which control robot actions into executable "scripts". For each application, an inventory of modules is provided based on task characteristics. A script's performance is evaluated using off-line simulation of the application robot and environment. Successful scripts for a specific task are found using the genetic algorithm search technique. The thesis discusses the implementation of this approach. The approach is applied to an example restoration task aboard the USS Constitution. The results of this application demonstrate that the scripts can be trained to succeed over the range of characteristic field environment variability. Also, one action module inventory may generate successful scripts for various robot configurations. Furthermore, successful scripts can be generated more quickly by including action modules which solve taskspecific problems. Supervisor: Dr. Steven Dubowsky Title: Professor of Mechanical Engineering