Computation of Driving Pleasure based on Driver's Learning Process Simulation by Reinforcement Learning 2013-01-0056
In order to improve the driver's experiences such as driving pleasure, it is important to evaluate the relationship between various vehicle characteristics and the driver's feeling. Although methods such as sensory subjective evaluation are commonly used, the mechanism behind them is not yet fully understood.
In this paper we introduce a novel method for evaluating driving pleasure based on the numerical simulation of the driver's learning process. As an example of this method we evaluate the relationship between mechanical property of steering system and pleasure felt during the driver's learning process.
One possible method to simulate the driver's learning process is machine learning. Reinforcement learning has been studied for simulating the human's brain function to learn . We use machine learning to create the reinforcement learning driver model, and a simple vehicle simulation model which are combined as a human-vehicle model. Then the model, with four different settings of steering stiffness, is simulated to learn to drive on a winding road constructed with two curves. The result shows that the characteristics of driver's learning process depend on the steering stiffness. We also find that there is a trade-off between the learning speed at the beginning and the learned level at the end of the learning process. So we estimate there is an optimal steering stiffness for continuous progress while learning how to drive, with which the driver can feel a high sense of accomplishment.
The aim of this research is to investigate whether the driver's progress process can be simulated or not. So in this study, we used the simple vehicle and driver model. We will continue to develop more precise models of both vehicle and driver to unearth the mechanisms of driving pleasure.