Force feedback controller of a parallel haptic device via online adaptive dynamic programming

Abstract

Haptic devices can replicate real-time force signals, which demand high accuracy and real-time performance. However, several common factors affect the accuracy of system modeling, which potentially limit the performance of haptic devices. These factors include unavoidable errors and unmodeled dynamics in the systems themselves, as well as uncertainties stemming from human–computer interaction (HCI) processes. Considering these unfavorable circumstances, this paper designs and fabricates a novel parallel haptic device with three degrees of freedom (DoF). To enhance modeling accuracy, a model-data driven approach is employed, utilizing the theoretical dynamic model as prior knowledge. A recurrent neural network (RNN) is used to approximate the actual system model. Subsequently, the online adaptive dynamic programming (ADP) technique is applied to achieve force reproduction. Critic-action networks are established to minimize the search for the established value function by continuously updating the weights of both networks in real-time. Through rigorous experimentation, the proposed haptic device successfully reproduces the desired force sensing signal and exhibits excellent adaptability.