Neural network hybrid position/force control

The authors extend the application of a multilayered feedforward network to the hybrid position/force control problem. Using the measured positions and forces during an assembly task as inputs to a neural network, the necessary selection matrix and artificial constraints can be computed by the network. The authors use the peg-in-the-hole insertion problem to demonstrate their method. The neural network hybrid position/force controller is shown to correctly switch to the required position and force control modes and to recall the desired positions and forces required for each subcontrol task.