Spatial Generalization of Optimal Control for Robot Manipulators

摘要

A diffusion-based learning approach is presented to generalize optimal control of a robot manipulator over a bounded workspace. By assuming that, for some sets of initial and desired terminal conditions of the robot's positions, we already have the numerical optimal control inputs (for example, by using some complex numerical calculation techniques), this approach first uses radial basis function (RBF) network to parameterize these control inputs by a set of weight matrixes. Diffusion-based algorithm is then applied to generalize these weight matrixes for different terminal position conditions. This approach greatly reduced calculation cost for the robot to find its optimal control. Diffusion-based algorithm is a parallel distributed learning approach, it only requires local interaction between the nodes of a learning network (a lattice) and can be realized by resent IC hardware technology easily. Computer simulations of a 2 D.O.F. planner robot arm show the effectiveness of this approach.