NEURAL-NETWORK PATH-PLANNING AND DIGITAL ADAPTIVE CONTROL OFREDUNDANT ROBOTS

RESEARCH WORK ON MANIPULATOR CONTROL HAS BEEN CONCENTRATED ON DEVELOPING CONTINUOUS-TIME CENTRALIZED CONTROL SCHEMES. IN PRACTICE, ROBOTS ARE CONTROLLED BY DIGITAL COMPUTERS AND THE SOLUTION OF THE CONTROL PROBLEM BASED ON THE ASSUMPTION OF CONTINUOUS TIME OPERATION CAN LEAD TO DEGRADATION OF PERFORMANCE AND POSSIBLE INSTABILITY OF THE OVERALL SYSTEM. FURTHERMORE, CENTRALIZED ROBOT CONTROLLERS REQUIRE EXTENSIVECOMPUTER MEMORY AND TIME, SINCE COMPUTATION OF A JOINT TORQUE IS BASED ON THE PROCESSING OF INFORMATION FROM ALL JOINTS. A CENTRALLY CONTROLLED ROBOT IS NOT FAIL-SAFE SINCEIT CAN LOSE CONTROL OF ALL JOINTS IF ONLY ONE JOINT SENSOR IS IN ERROR. THE MAIN OBJECTIVE OF NETROLOGIC'S PROPOSED RESEARCH IS TO DEVELOP A DISCRETE-TIME DECENTRALIZED ADAPTIVE CONTROL SCHEME FOR REAL-TIME DIGITAL CONTROL OF HIGH PERFORMANCE ROBOT MANIPULATORS, AND TO EXPERIMENTALLY DEMONSTRATE ITS PERFORMANCE. SUCH SCHEMES ARE NEEDED SINCE DECENTRALIZED (INDEPENDENT JOINT) CONTROL IS COMPUTATIONALLYEFFICIENT AND IS SUITABLE FOR PARALLEL PROCESSING WITHIN A DISTRIBUTED COMPUTER ARCHITECTURE. THE FIELDS ADAPTIVE CONTROL AND ADAPTIVE NEURAL NETWORKS, HAVE BEEN DEVELOPING INDEPENDENTLY BUT HAVE THE ADAPTATION MECHANISM IN COMMON. NETROLOGIC WILL ALSO INVESTIGATE RECENT PROGRESS IN NEURAL NETWORK APPLIED TO ROBOTIC CONTROL PROBLEMS. IN PARTICULAR,THE APPLICABILITY OF NEURAL NETWORKS CONCEPTS TO ADAPTIVE CONTROL OF ROBOT MANIPULATORS WILL BE EXPLORED. THE PROPOSED WORK CONSISTS OF TWO PHASES. PHASE 1 IS OF 6 MONTHS DURATION AND WILL BE DEVOTED TO THE DEVELOPMENT OF CONTROL SCHEME, COMPUTER SIMULATIC AND A FEASIBILITY STUDY OF ADAPTIVE NEURAL NETWORK FOR ROBOT CONTROL. IN PHASE II, NETROLOGIC PROPOSES TO IMPLEMENT THE CONTROL SCHEMES ON A PUMA 562 ARM AND TO DEMONSTRATE ITS PERFORMANCE.