An approach to tune fuzzy controllers based on reinforcement learning for autonomous vehicle control

In this paper, we suggest a new approach for tuning parameters of fuzzy controllers based on reinforcement learning. The architecture of the proposed approach is comprised of a Q estimator network (QEN) and a Takagi-Sugeno-type fuzzy inference system (TSK-FIS). Unlike other fuzzy Q-learning approaches that select an optimal action based on finite discrete actions, the proposed controller obtains the control output directly from TSK-FIS. With the proposed architecture, the learning algorithms for all the parameters of the QEN and the FIS are developed based on the temporal-difference (TD) methods as well as the gradient-descent algorithm. The performance of the proposed design technique is illustrated by simulation studies of a vehicle longitudinal-control system.