Injecting Conflict Situations in Autonomous Driving Simulation using CARLA
Tsvetomila Mihaylova, Stefan Reitmann, Elin A. Topp, Ville Kyrki
TL;DR
This paper addresses the lack of accessible, reproducible data on conflict situations in autonomous driving, particularly in shared autonomy where takeover decisions are necessary. It introduces a CARLA-based toolkit that injects controllable conflict scenarios, selecting scenarios using a framework that prioritizes urgency and driver response, and demonstrates these conflicts on both default CARLA maps and a custom OpenDRIVE/OpenScenario map. The contributions include a set of implemented conflict types (e.g., vanishing lane markings, weather effects, narrowing roads, obstacles, sensor noise) and a simple but extensible controller plus basic situation-awareness features, along with setup and loading instructions for scenarios and custom maps. The toolkit enables human-in-the-loop studies of explanations and situation awareness, supporting reproducible experiments and providing a foundation for safety metrics, regulatory considerations, and future HRI research in autonomous driving.
Abstract
Simulation of conflict situations for autonomous driving research is crucial for understanding and managing interactions between Automated Vehicles (AVs) and human drivers. This paper presents a set of exemplary conflict scenarios in CARLA that arise in shared autonomy settings, where both AVs and human drivers must navigate complex traffic environments. We explore various conflict situations, focusing on the impact of driver behavior and decision-making processes on overall traffic safety and efficiency. We build a simple extendable toolkit for situation awareness research, in which the implemented conflicts can be demonstrated.