Neural force control (NFC) for complex manipulator tasks

In order to apply intelligent robot control to complex force/position tasks, we developed a novel concept for force/position control based on neural networks (NN). A neural dynamic net (NDN) contains a neural computed torque controller, which delivers a precise mapping of the inverse model of the real manipulator. Specifically, the mapping considers different types of nonlinear properties, which are essential for this type of control, but are hard to model analytically. Furthermore the inverse kinematics is represented by a neural kinematics network (NKN), which includes strategies for avoidance of singularities, self-collisions, and conflicts with workspace constraints. This neural control approach has been tested in simulations and will be applied to a 6 DOF industrial manipulator to various demanding tasks including screw removal and surface tracking with constant normal force.