Development of neurofuzzy controllers for failure tolerance of kinematically redundant robotic manipulators

Since the introduction of industrial manipulators, ensuring safe and robust robotic operations has been a great concern when humans and robots are working in the same area. A failure tolerant robotic control system reduces the risk of inadvertent motion following a failure that could injure a person or damage the surrounding environment, and increases the certainty that the task is completed according to the original plan.;For the successful development of this controller, four key problems in robotic failure tolerance are addressed. First, the difficulties in determining a solution of the inverse kinematics is considered. As a redundant manipulator has infinitely many possible joint positions, a scheme is developed to determine an optimal solution that can incorporate failure information. Secondly, nonlinear manipulator dynamics are addressed through the development of a new fuzzy proportional, integral, and derivative (PID) controller. This controller shows robustness for controlling nonlinear systems which are not controllable by conventional PID controllers. The third area is the determination of the friction encountered in arm operation. A real-time adaptive B-Spline Neural Network is used to successfully model nonlinear friction as functions of both velocity and position changes. Finally, an effective sensor failure identification technique is addressed through the application of an output observer with a flexible residual threshold calculation using fuzzy logic. Included in this work is an example of the controller implemented on a simulated Robotics Research Corporation K-1207i seven degrees of freedom industrial manipulator.;In this dissertation, a failure tolerant control system is developed for redundant manipulators. The systems is designed to operate with both sensor and joint component failures. A real-time controller is developed which automatically utilizes the extra kinematic capacity in a redundant manipulator to compensate for failed components, quickly minimizes deviations in the arm trajectory, adapts to a certain degree of unexpected variation in the operating environment, and ascertains when the sensor data indicates a significant failure.