Learning to Drive using Inverse Reinforcement Learning and Deep Q-Networks

TL;DR

This paper introduces a novel IRL method using Deep Q-Networks for autonomous driving, enabling the extraction of reward functions in complex environments, leading to collision-free and human-like driving behaviors.

Contribution

It presents a new IRL approach with Deep Q-Networks tailored for large state spaces in autonomous driving scenarios, demonstrating effective reward learning and realistic driving actions.

Findings

The method accurately infers reward functions from sensor data.

The agent achieves collision-free navigation after few learning iterations.

The driving behavior resembles human lane changing patterns.

Abstract

We propose an inverse reinforcement learning (IRL) approach using Deep Q-Networks to extract the rewards in problems with large state spaces. We evaluate the performance of this approach in a simulation-based autonomous driving scenario. Our results resemble the intuitive relation between the reward function and readings of distance sensors mounted at different poses on the car. We also show that, after a few learning rounds, our simulated agent generates collision-free motions and performs human-like lane change behaviour.