Multiple state estimation reinforcement learning for driving model: driver model of automobile

In this paper, a multiple state estimation method for reinforcement learning model was proposed. The steering manoeuvre of a vehicle is learned from a reward. The reward was evaluated whether the vehicle is on the road or not. The reward and control signal were calculated by multiple state using a vehicle dynamics model. From the simulation result, this model can drive on unknown road configuration or velocity condition. This model also explains the gaze control by changing information using a control policy or an environment.