NEURAL NETWORKS FOR AUTONOMOUS MOTOR CONTROL

CURRENT METHODS IN MOTOR CONTROL HAVE PROBLEMS DEALING EFFECTIVELY WITH HIGHLY VARIABLE ENVIRONMENTS AND SENSORY-MOTOR PARAMETERS. THE PROPOSED WORK OVERCOMES SOME OF THESE DIFFICULTIES BY BUILDING A NEURAL CONTROLLER THAT LEARNS ADAPTIVE MOTOR CONTROL FROM ITS OWN EXPERIENCE. THE OBJECTIVE OF THE PROPOSED PHASE I STUDY IS TO IMPLEMENT A SINGLE-JOINTED ARM AND CONTROLLER FOR POSITIONING UNFORESEEN PAYLOADS WITH ACCURATE AND STABLE MOVEMENTS. THE PROPOSED IMPLEMENTATION WILL BE BASED ON A WORKING COMPUTER SIMULATION THAT HAS BEEN SHOWN TO ACHIEVE AUTONOMOUS ADAPTIVE CONTROL. THE NEURAL ARM HAS BEEN DESIGNED TO ADAPTIVELY CONTROL ANY NUMBER OF SENSORY INPUTS WITH LINKS OF ANY NUMER OF JOINTS. THE FEEDFORWARD NATURE OF CONTROL WILL ALLOW PARALLEL IMPLEMENTATION IN REAL TIME CROSS MULTIPLE JOINTS. IT WILL TOLERATE INTERNAL NOISE, PARTIAL SYSTEM DAMAGE AND CHANGES IN THE MECHANICAL AND SENSORY PARAMETERS OF THE ROBOT AS THEY OCCUR OVER TIME. THIS ADAPTABILITY ELIMINATES THE NEED FOR OPERATOR CALIBRATION. IN PHASE II OF THIS PROJECT, THIS NEURALL CONTROLLER WILL BE EXTENDED TO MULTIPLE JOINTS.