Interaction-awareness based Intention Inference of Lag Vehicle in Lane Changing Decision-Making Process for Autonomous Driving

Abstract

Cooperative driving behavior during lane-changing decision-making processes is expected to improve the safety of autonomous vehicles. However, since the intention of the lag vehicle driver is changeable and cannot be directly observed, the challenge of achieving cooperative driving behavior still remains. In this paper, an intention inference method that combines interaction-awareness information is proposed to infer the uncertain intention of the lag vehicle driver from multiple time-series driving data. The method is based on the Hidden Markov Model with the Gaussian Mixture Model (GMM-HMM) structure, which could enhance the inference performance by catching the feature that human drivers’ intentions cannot change instantaneously. Furthermore, variables are selected to train the proposed model based on the decision-making mechanism of both drivers with the interaction during lane-changing processes. High dimensional training data is alleviated by using the virtual collision point which could convert numerous training variables into an associative variable. Experimental results demonstrate that the proposed method can improve the inference performance and the collision avoidance performance in lane-change scenes compared with previous methods.