Personalized Human-Machine Cooperative Lane-change Based on Recurrent Neural Network

To reduce the interference and conflict of human-machine cooperative control and improve the friendliness and acceptability of intelligent vehicles, a personalized human-machine cooperative lane-change method was proposed. Firstly, a test platform with active steering in the loop was built based on MATLAB/Simulink and PreScan. The lane-change test condition for data collection were designed, and the lane-change behaviors data of five drivers with different personalities were collected. Then, a multi-layer recurrent neural network (RNN) was built based on Long Short-Term Memory (LSTM) to learn the lane-change behaviors of each driver and a personalized lane-change method was trained for each driver. In addition, the lane-change data from previous research that did not include these five drivers was used to retrain the RNN, and this RNN was used as the default lane-change method for comparison. The training results show that compared with the default method, the personalized method is closer to the driver's steering wheel operation in the lane-change process, and the more stable the driver's driving behavior is, the better the fitting effect will be. Finally, the personalized lane-change method and the default one are respectively embedded into the machine system and the steering wheel control was connected to the machine system, and then the typical drivers were invited to conduct the personalized human-machine cooperative lane-change test. The test results show that the machine control method without personalization is easy to interfere and conflict with the driver, and the personalized human-machine cooperation can effectively reduce the human-machine conflict and lighten the driver's operating load. The more special the driver's driving behavior is, the better the effect of personalized man-machine cooperative will be.