Tri-Projection Neural Network for Redundant Manipulators

Resolving redundancy for manipulators subject to various limits is significant in robotics. The problem is often handled by an optimization formulation with joint velocity being the decision variable, for which how to address the joint acceleration constraint is challenging. Motivated by this problem, a dynamic neural network with triple projections, called tri-projection neural network (TPNN), is developed for quadratic programs with a constraint on the state evolution of the neuron states. The proposed TPNN is applied to resolving redundancy of an ABB IRB 140 industrial manipulator with velocity inputs subject to joint acceleration constraints. Simulation comparisons with an existing method demonstrate the superiority of the developed TPNN in fully employing the acceleration capability of the manipulator.