Learning to Navigate Intersections with Unsupervised Driver Trait Inference

TL;DR

This paper introduces an unsupervised approach using variational autoencoders and reinforcement learning to infer driver traits from trajectories, improving autonomous intersection navigation by adapting to different driver behaviors.

Contribution

It presents a novel unsupervised trait inference method combined with reinforcement learning for autonomous driving in intersections, without requiring labeled trait data.

Findings

Outperforms state-of-the-art baselines in T-intersection navigation

Effectively infers driver traits from trajectories without ground truth labels

Enhances safety and efficiency by adapting to different driver behaviors

Abstract

Navigation through uncontrolled intersections is one of the key challenges for autonomous vehicles. Identifying the subtle differences in hidden traits of other drivers can bring significant benefits when navigating in such environments. We propose an unsupervised method for inferring driver traits such as driving styles from observed vehicle trajectories. We use a variational autoencoder with recurrent neural networks to learn a latent representation of traits without any ground truth trait labels. Then, we use this trait representation to learn a policy for an autonomous vehicle to navigate through a T-intersection with deep reinforcement learning. Our pipeline enables the autonomous vehicle to adjust its actions when dealing with drivers of different traits to ensure safety and efficiency. Our method demonstrates promising performance and outperforms state-of-the-art baselines in the T-intersection scenario.