Automatic expert system for fuzzy control of robot trajectory in joint space

Trajectory control for manipulation robots in joint space is used both in industrial and research applications. This paper shows how an expert system can be developed to control a manipulation robot and to generate a trajectory for one single mobile robot joint, supposing all the other joints are locked in fixed positions. The automatic expert system includes a control system for trajectory tracking to simulate a joint trajectory with the provided data. The expert system learning module uses adaptive neuro-fuzzy learning techniques, based on the gradient descent and the least mean squares methods. The learning module improves the expert system performances to optimally choose the trajectory type. The experimental tests for the prototype expert system prove the efficiency of the neuro-fuzzy rules for trajectory selection. The performances of the expert system are also determined by the heuristic rules used to modify the closed loop gains used in the control system to simulate the robot response and to generate the trajectory. The efficiency of the joint trajectory control is shown by making a comparative analysis between the obtained trajectories and those simulated as a response of the control system.