A personalized lane keeping system for autonomous vehicles based on recurrent neural network with temporal dependencies

This paper presents a personalized lane keeping system for an autonomous vehicle using recurrent neural network (RNN) with long short term memory (LSTM) cell. The proposed algorithm is trained by datasets collected by manual driving of three drivers. The collected driving data is analyzed for the target lane offset and responsiveness to the road curvature of each driver. 178744 and 76605 datasets are used to train and validate the LSTM-RNN based model. An encoder is used to standardize the input feature to improve the accuracy of network training. The proposed lane keeping algorithm for each driver has been evaluated through prediction accuracy analysis and simulation study using MATLAB/Simulink and Carsim. 99.7 % of the prediction error of steering wheel angle was bounded between −0.87 deg to 0.89 deg with mean of 0.01 deg. The simulation results show that the proposed algorithm precisely modeled the lane keeping characteristics of three drivers.