A hybrid numerical/knowledge-based system for locomotion control of a multilegged articulated robot.

摘要

Goal-directed interaction of articulated bodies (such as humans, high animal species, and robots) with the environment has long been a problem area for control engineers, mechanical engineers, biomechanical scientists, and behavioral psychologists. Nowadays it is a major problem area in the fields of robotics and graphical animation, where researchers seek to mimic purposeful movements and motion.;This thesis proposes a model for understanding, synthesizing, and learning coordinated movements within a cohesive framework which combines the mathematical rigor of approaches taken in robotics, biomechanics, and artificial intelligence literature with the behavioral relevance of psychological approaches. The model is demonstrated in a locomotion control system for a four-legged articulated robot.;One of the primary objectives of this research is to examine the effectiveness of merging the concepts of computer numerical control (mathematical rigor of approaches taken in robotics, and biomechanics) with the concepts of motor learning (knowledge-based processing as studied in artificial intelligence and behavioral psychology). The expected result is that both methodologies can act together to provide an autonomous motion control system that is capable of synthesizing and learning coordinated locomotive movements and improving its performance as a result of practice and experience.;The proposed locomotion control system currently navigates a four-legged articulated robot through a simulated environment containing obstacles, holes, inclines, rough terrain, etc. The equations of motion of the robot are solved by a recursive method, implemented on an IRIS-2400 graphics workstation. The robot must show some prudence in choosing the most appropriate locomotive skill at any point during its navigation: to slow down while turning a corner, to prefer paths it has traveled on before, to reduce its total energy consumption in executing a mission, etc.;However, none of these fields yet offers any satisfactory solution to the problem of understanding, synthesizing, and learning coordinating movements. There are partial solutions in psychological studies, graphical animation systems, and robotics simulations. It seems the lack of a well-defined methodological framework for analysis of movements has been a major impediment to both experimental simulations and theoretical studies.;The emphasis in this research is twofold: (1) acquiring high agility through a learning process, as might be required to quickly move from one point to another on a structure under construction, or needing repairs in an emergency; (2) enabling the robot to solve, 'on-board' and in real-time, complex problems of manipulator dynamics required for locomotion control.