Study Of Human-like Steering Control Driver Model For Intelligent Vehicle Based On Trajectory Similarity

Driver model is one of key components for realizing autonomous driving of intelligent vehicles,and curved road driving is an important scenario for driving safety and autonomous driving function verification.To make intelligent vehicles and traditional vehicles share road together harmoniously,it is necessary to study the trajectories of experienced human drivers on curved road and let intelligent vehicles obtain a similar steering control characteristic as human drivers'.In this paper,based on the analysis towards the trajectories data from experienced human drivers,a humanlike steering characteristic evaluation method is proposed.Also,a human-like steering control driver model as well as its optimization frame are proposed,which have been verified by simulation and experiment results.The paper is organized as follows:Firstly,based on high-resolution locating device,driving data from several experienced human drivers were collected in a two-lane urban curved test field under different speeds.To compare trajectories from different experienced drivers,the coordinates updated by time are transferred into the lateral deviation sequence updated by virtual landmarks.Based on the analysis towards the similarity between experienced drivers,the target trajectory which would be used in driver model optimization is proposed as well as a human-like steering characteristic evaluation method.Secondly,based on the analysis of drivers' visual mechanism when passing a curved road,a preview decision module is proposed using fuzzy inference system,which can adaptively adjust the preview point in both lateral and longitudinal direction.Combing this decision module with the optimal preview driver model,a human-like steering control driver model is proposed.In order to optimize the fuzzy laws in the preview decision module,an off-line as well as an on-line driver model optimization frame are proposed using modified ACO algorithm and fuzzy reinforcement learning method respectively.Thirdly,a same simulation environment as the test field was built in PreScan/Simulink platform and the driver model as well as optimization frame were simulated in Matlab/Simulink software.The simulation results show that in each speed,the proposed driver model has a good human-like steering characteristic.Additionally,both optimization frame can improve the human-like steering characteristic of the driver model,however the off-line frame has a better optimization results and the on-line frame has a better optimization efficiency.In addition,simulation results verify that the proposed human-like driver model is able to be applied on different curvature roads.At last,the feasibility of the proposed driver model is verified in the steering angle tracking experiment using a real vehicle,and the experiment results show that the steering angle sequence generated by the driver model can be realized by the EPS steering system on a normal vehicle.The effectiveness of the proposed drirver model is verified in the indoor path tracking experiment based on the indoor locating device and model vehicle,which proves that the driver model can realize the path tracking task in a real driving environment.