Design and development of a four degree-of-freedom robot arm and multi-neural network intelligent path-planning subsystem

This thesis concerns a novel path-planning subsystem applied in the intelligent control of a four degree of freedom robot arm. The "intelligent control" is a multi-neural network based control program written in C. The control algorithm uses two Adaptive Resonance Theory II (ART2) neural network classifiers and a Bi-directional Associative Memory System (BAM System). These neural networks allow the arm to remember the position of obstacles in its workspace, allowing the generation of trajectories (around the obstacles) in future movements. Success in this project is considered to have several levels. Because the robot arm is being designed and built as part of this project, the first level of success is the completion of, and full dynamic control of the robot arm. The second level of success is the successful completion of the controlling program. This includes the reliable operation of the ART2 and BAM System neural networks. The final level of success is the actual performance of the arm/intelligent controlling program in mapping and avoiding obstacles in the robot's work area. The controlling algorithm is considered a 'subsystem' which could be used with other navigational sensory and control subsystems in a variety of applications. The idea of multiple subsystems available in an overall application provides the protection of back-up systems in case of primary control system failure, or inability to detect obstacles in certain situations. Information in this thesis includes some neural network theory, path-planning in robotic control/trajectory generation methods, dynamic control system theory concerning the hardware and software used in this system, this system's approach at providing trajectory generation and obstacle avoidance, and implementation results.