eduzhai > Applied Sciences > Engineering >

Development of A Stochastic Traffic Environment with Generative Time-Series Models for Improving Generalization Capabilities of Autonomous Driving Agents

  • Save

... pages left unread,continue reading

Document pages: 7 pages

Abstract: Automated lane changing is a critical feature for advanced autonomous drivingsystems. In recent years, reinforcement learning (RL) algorithms trained ontraffic simulators yielded successful results in computing lane changingpolicies that strike a balance between safety, agility and compensating fortraffic uncertainty. However, many RL algorithms exhibit simulator bias andpolicies trained on simple simulators do not generalize well to realistictraffic scenarios. In this work, we develop a data driven traffic simulator bytraining a generative adverserial network (GAN) on real life trajectory data.The simulator generates randomized trajectories that resembles real lifetraffic interactions between vehicles, which enables training the RL agent onmuch richer and realistic scenarios. We demonstrate through simulations that RLagents that are trained on GAN-based traffic simulator has strongergeneralization capabilities compared to RL agents trained on simple rule-drivensimulators.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...