Abstract: The gradual increase in the level of automatic driving means that the power of driving execution is gradually transferred from the driver to the vehicle system, and the responsibilities of the driver also change accordingly. A large number of studies have shown that the attention span of the driver of an autonomous vehicle is closely related to driving safety, and there are differences in the driver state thresholds required for different levels of autonomous driving. This research proposes an long short term memory(LSTM) network driver state prediction model(LSTM-DSDM) that integrates the driver state discrimination mechanism to realize the prediction of the driver's load state and the recognition of the state transition stage, and based on the task requirements of drivers under different levels of automatic driving, a driver's load status monitoring strategy of low-level recognition, high-level prediction is proposed. The experimental results show that the driver's load status monitoring strategy proposed in this study can effectively respond to the driver's load status monitoring needs of different autonomous driving levels. The driver's load state prediction model built in this study has a high recognition rate under the condition of low autopilot level, which can reach more than 90%; under the condition of high autopilot level, the prediction rate of the model can achieve the prediction effect to a certain extent, and it can also be used to study the transition phase of the driver's load state.

Key words: driving automation level, driver capability requirements, driver condition monitoring, load transition, long short term memory(LSTM)