Predicting vehicle trajectories with inverse reinforcement learning

Sammanfattning: Autonomous driving in urban environments is challenging because there are many agents located in the environment all with their own individual agendas. With accurate motion prediction of surrounding agents in the environment, autonomous vehicles can plan for more intelligent behaviors to achieve specified objectives, instead of acting in a purely reactive way. The objective of this master thesis is to predict the future states of vehicles in a road network. A machine learning method is developed for trajectory prediction that consists of two steps: the first step is an inverse reinforcement learning algorithm that determines the reward function corresponding to an expert driver behavior extracted from real world driving, the second step is a deep reinforcement learning module that associates high level policies based on vehicles observations. Regular drivers take into account many factors while making tactical driving decisions, which cannot always be represented by the conventional rule-based models. In this work, a novel approach to learn the driver behavior by extracting suitable features from the training dataset is proposed. The accuracy of predictions is evaluated using the NGSIM I-80 dataset. The results show that this framework outperforms a constant velocity model when predicting further than 6 seconds into the future.