The authors consider the problem of a robot manipulator operating in a noisy workspace. The robot is assigned the task of moving from an initial position P//i to a final position P//f. Since P//i this position can be known fairly accurately. However, since P//f is usually obtained as a result of a sensing operation, possible vision sensing, the authors assume that P//f is noisy. The authors propose a solution to achieve the motion which involves a learning automaton, called the discretized linear reward-penalty (DL//R//P) automaton. The strategy proposed does not involve the computation of any inverse kinematics. Alternatively, an automaton is positioned at each joint of the robot, and by processing repeated noisy observations of P//f the automata operate in parallel to control the motion of the manipulator. The advantages and the possible disadvantages of the scheme are also discussed.
