Torque sensorless decentralized neuro-optimal control for modular and reconfigurable robots with uncertain environments

Abstract

A technical challenge of addressing the decentralized optimal control problem for modular and reconfigurable robots (MRRs) during environmental contacts is associated with optimal compensation of the uncertain contact force without using force/torque sensors. In this paper, a decentralized control approach is presented for torque sensorless MRRs in contact with uncertain environment via an adaptive dynamic programming (ADP)-based neuro-optimal compensation strategy. The dynamic model of the MRRs is formulated based on a novel joint torque estimation method, which is deployed for each joint model, and the joint dynamic information is utilized effectively to design the feedback controllers, thus making the decentralized optimal control problem of the environmental contacted MRR systems be formulated as an optimal compensation issue of model uncertainty. By using the ADP method, a local online policy iteration algorithm is employed to solve the Hamilton–Jacobi–Bellman (HJB) equation with a modified cost function, which is approximated by constructing a critic neural network, and then the approximate optimal control policy can be derived. The asymptotic stability of the closed-loop MRR system is proved by using the Lyapunov theory. At last, simulations and experiments are performed to verify the effectiveness of the proposed method.