Design of Longitudinal Control for Autonomous Vehicles based on Interactive Intention Inference of Surrounding Vehicle Behavior Using Long Short-Term Memory

Abstract

This paper presents a method of intention inference of surrounding vehicles' behavior and longitudinal control for autonomous vehicles. A Recurrent Neural Network (RNN) based on Long Short-Term Memory (LSTM) cells has been used to predict the future driving lane of surrounding vehicles. Interaction among the adjacent vehicles is considered in the RNN to improve the behavior prediction accuracy. A Model Predictive Control (MPC) has been designed to derive the longitudinal control input of the autonomous vehicle in a predictive manner based on the prediction results. The proposed behavior prediction algorithm has been evaluated according to its behavior classification accuracy. Also, the longitudinal control algorithm has been validated in car-following scenarios with the existence of cut-in vehicles via computer simulations. Experimental results show that the proposed predictor improves the performance of behavior prediction and the longitudinal control method enables autonomous vehicles to maintain safety with respect to the cut-in vehicles with proper ride quality.