Obstacle Avoidance Based Anfis and Tracking Control a Redundant Robot Manipulator

This paper addresses the issue of obstacle avoidance of a robot manipulator fixed in a predefined environment. This is achieved by using a selfmotion technique related to the robot arms. It is obtained by projecting the gradient of a scalar function (representing the criterion) on the null space of the Jacobean matrix related to the robot. Two important analytical methods are generally used: Gradient h method, when h is computed using Euclidean distance or pseudo-distance, and task priority method. These methods are tested when a redundant robot with 3 degrees of freedom (DOF) tracks a Cartesian trajectory in the presence of obstacles. After that, an input/output data file is constituted by selecting, at each instant, the self-motion data related to the obstacle avoidance method giving the best performance among the three analytical methods. An adaptive neuro fuzzy inference system (ANFIS) based Takagi-Sugeno model is elaborated in order to optimize performance and to simplify the self-motion computation. The network learning is carried out in order to copy the input/output data selected previously for the self-motion computation. The trajectory tracking control incorporating obstacle avoidance where the self-motion is generated by the trained ANFIS led to better performances (actuators torques and tracking control errors are weaker) than those obtained by each the analytical methods. These results show that it is possible to generate self-motion through an ANFIS network resulting from an adequate mixture between the self-motion data coming from several analytical methods or others.