Neural network support for simultaneous optimization of torque and deviation in robot path planning

A novel architecture of the robot's trajectory planner is proposed. It uses both a conventional trajectory generation algorithm based on underdetermined splines optimized with respect to some quadratic criteria together with a neural network module capable of coping with a nonlinear multi-objective design problem. The backpropagation network structure is chosen. The network has learned the positions of trajectory via-points that fit the requirements concerning both the joint driving torques and deviation from nominal trajectory. The simulation study shows the feasibility of the proposed solution in the cases of one and two via-points.<
>