Application of reinforcement learning control to a nonlineardexterous robot

摘要

In this paper, the effects of basic parameters in reinforcementlearning control such as eligibility, action and critic network weights,system nonlinearities, gradient information, state-space partitioning,variance of exploration were studied in detail. We attempt to increasefeasibility for practical applications, implementation, learningefficiency, and performance. Reinforcement learning is then applied forcontrol of a nonlinear dexterous robot. This control problem dictatesthat the learning is performed online, based on binary and real valuedreinforcement signal from a critic network, without knowing the systemmodel nonlinearity. The learning algorithm consists of an action andcritic networks that learn to keep the multifinger hand of the dexterousrobot within desired limits