A conceptual framework of decentralized learning neural network control approach for multi-robot cooperation in an object balancing task

摘要

This paper presents a conceptual framework of a neural network control approach for robot manipulator cooperative, which is based on decentralized learning. Back propagation neural network is used for learning procedure to adapt and adjust the neuron-controller's parameters which depend on the approximated error. Dynamic model of two cooperating 3-DOF robot manipulators are defined and implement with neural network control. Visual feedback enables two robots to correct and calibrate their movement to compensate their object balancing task whereas both robots hold a flat plate balancing a round object on it. This conceptual framework of the decentralized learning procedure will be verified by a simulation and experiments in near future.