CARLA Real
Traffic Scenarios

Overview

Our work introduces interactive traffic scenarios in the CARLA simulator, which are based on real-world traffic. We concentrate on tactical tasks lasting several seconds, which are especially challenging for current control methods. The CARLA Real Traffic Scenarios (CRTS) is intended to be a training and testing ground for autonomous driving systems. To this end, we open-source the code under a permissive license (source code) and present a set of baseline policies.

CRTS combines the realism of traffic scenarios and the flexibility of simulation. We use it to train agents using a reinforcement learning algorithm. We show how to obtain competitive polices and evaluate experimentally how observation types and reward schemes affect the training process and the resulting agent's behavior.

Experimentally, we find that bird’s-eye view and dense rewards combined with a penalty for a failure to complete an episode generalizes best to validation scenarios.

CRTS is described in a paper, which is currently under review for CORL.

teaser.mp4

Policies analysis

CRTS can be used to evaluate various details of the vehicle decision stack. In our experiments, we concentrate on observation types and reward schemes. We concluded that bird’s-eye view and dense rewards perform best.

Below we present a more detailed analysis along with the most interesting episodes recorded during agent evaluation.

Contributions

We intend CRTS to be used by the research community. In particular, we encourage submissions of evaluations of other polices on CRTS. This can be done with the form below or contacting us directly. Feel free to email us also regarding other research questions. In particular, we would be happy to incorporate other datasets, metrics and we are open to external contributions.