A Study of Neural Network Based Inverse Kinematics Solution for a Planar Three Joint Robot with Obstacle Avoidance

A neural network based, collision–free, inverse kinematics solution of a three joint robotic manipulator is presented in this paper. An important application of it is the automatic planning, regarding the motion and final orientation of the links, for a given desired end point in a workspace having one or more obstacles of varied sizes at any orientation and at any point. Inverse kinematics problem is generally more complex for robotic manipulators.For simplicity the motion of the robotic links is chosen to be planar in x-y workspace rendering it redundant. First, we have to determine a set of manipulator joint angle values for given end points and obstacle size and location, such that all the three links avoid the obstacle and at the same time summation of the distance of the 3-links from the obstacle is maximum. Our objective function is the sum of the distances of the obstacle with each of the links of the three link robot manipulator. Our equality and inequality constraints ensure the obstacle avoidance and at the same time confirm that the robot reaches the destination point in the workspace. We use the pseudo-obstacles concept for expanding the object to be moved to a point and shrinking the obstacles to a point. The path-planning problem is reduced to finding a safe path for the expanded links of the robot avoiding a set of the reduced obstacles which are nothing more than points now. If this path is collision free then the entire robot is collision free for the original case. Then all of these joint