Injecting Conflict Situations in Autonomous Driving Simulation using CARLA

Tsvetomila Mihaylova, Stefan Reitmann, Elin A. Topp, Ville Kyrki · 2025

Read Injecting Conflict Situations in Autonomous Driving Simulation using CARLA by Tsvetomila Mihaylova, Stefan Reitmann, Elin A. Topp, Ville Kyrki (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-16476.txt.

Introduction

Tsvetomila Mihaylova and colleagues staged traffic conflicts between self-driving cars and human drivers in a 2025 paper. Shared roads force automated and human drivers to negotiate, and misunderstandings at that interface cause the riskiest moments. Using the CARLA driving simulator, the team built a set of exemplary conflict scenarios that arise under shared autonomy.

The scenarios probe how driver behavior and split-second decisions affect safety and traffic flow when both kinds of driver must react. The toolkit is deliberately simple and extendable, so other groups can add conflicts and rerun them. Situation awareness researchers gain a shared stage for testing how machines and people read each other's intentions.

Catalog record

CREATORTsvetomila Mihaylova, Stefan Reitmann, Elin A. Topp, Ville Kyrki
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE18 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-16476

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Injecting Conflict Situations in Autonomous Driving Simulation using CARLA Thanks: This work was partially supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Knut and Alice Wallenberg Foundation and Saab AB. Tsvetomila Mihaylova, Stefan Reitmann, Elin A. Topp, Ville Kyrki Abstract Simulation of conflict situations for autonomous driving research is crucial for understanding and managing interactions between av and human drivers. This paper presents a set of exemplary conflict scenarios in CARLA that arise in shared autonomy settings, where both av 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. I Summary In Level-3 automated driving [ 1 ] , the av operates on its own for most of the journey and the driver can concentrate on other tasks. In case of a conflict situation (i.e. a situation in which the autonomous system is uncertain how to proceed), the control has to be passed to the driver. In these cases, the system needs to provide explanations that would allow the human to understand the situation and act in a short time. In order to enable research in this scenario, we build a toolkit based on the CARLA simulator [ 2 ] , that allows HRI researchers to quickly simulate conflict situations and easily conduct user studies to explore how the human reacts to different explanations, which could lead to improvements in situational awareness. We implement several conflict types, chosen based on a framework for conflict simulation [ 3 ] . Our contributions are: • Implementation of conflicts situations in driving scenarios in the CARLA simulator, chosen based on their urgency and needed reaction time. • A simple extendable toolkit 1 1 1 https://github.com/aalto-intelligent-robotics/aisa-toolkit for experiments with situation awareness in autonomous driving using the CARLA simulator. • A basic setup that can showcase the implemented conflicts, which initiates a takeover request in case of uncertain lane marking detection. Fig. 1 : View from interior with detected lanes (green) and predicted path (purple). II Background In this section, we discuss details of the problem of shared autonomy in the context of automated vehicles to motivate why a simulation environment for conflict situations is needed and explain our reasoning behind choosing to implement specific conflicts. II-A Autonomous driving in shared autonomy In shared autonomy, automated systems and humans collaborate to execute tasks effectively [ 4 , 5 ] . When automated systems encounter limits defined by their design—such as environmental, locational, temporal, or traffic constraints, collectively termed odd ( odd ) [ 1 ] —a tor ( tor ) is triggered, switching to manual control. The system must detect conflicts [ 6 ] and provide sufficient context awareness [ 7 , 8 , 9 , 10 , 11 ] to help the human respond promptly. We focus on sae ( sae )-Level 3 (semi-autonomous driving), where the vehicle handles most of the journey while the driver can attend to other tasks [ 1 ] . We simulate the scenario where a human is in the automated vehicle and observes the driving situation. In case of uncertainty, caused by a conflict situation, a tor is initiated. II-B Conflicts in autonomous driving II-B 1 Data of conflict situation in autonomous driving Several works focus on simulating conflicts. Nair et al. predict situations requiring intervention and evaluate them in the CARLA simulator using static and dynamic obstacles and adverse weather conditions [ 12 ] . Haojie et al. simulate scenarios in the SUMO simulator, detecting conflicts based on vehicle motion [ 13 ] . Reitmann et al. develop a conflict simulation tool using Unity [ 14 ] . Datasets with conflicts from crash or near-crash situations are also available [ 15 , 16 , 17 ] . However, to the best of our knowledge, no open-source simulation meets these criteria: enabling easy conflict simulation, allowing human observation and control (supporting situation awareness research), and offering features to enhance situation awareness. II-B 2 Conflict classification The most general classification of conflicts can be seen as a binary classification within the problem of reaching the limit of the odd : first, the system reaches its technical limits, so it does not have the capabilities to find a solution for a given problem or has to break (traffic) rules. Second, the system’s technical capabilities are limited due to failure, e.g. sensor malfunctions. A lot of empirical studies have been conducted to analyse single types of conflicts within this binary split. A main concern in the class of system limits is lane markings (e.g. blurred/missing lane markings, secondary lane markings because of work zones, road curvatures) [ 18 , 19 , 20 ] . Other studies concentrated on the traffic dynamics and possible interactions with other participants, e.g. pedestrians, obstacles on the road, cut-in vehicles [ 21 , 22 , 23 , 24 ] or environmental factors [ 25 , 26 ] . A framework to build a common ground for singular conflicts in autonomous driving is proposed by Gold et al. , where conflicts are classified according to their urgency (time budget to solve the conflict), predictability (dependencies to other factors), criticality (safety risks) and driver response (complexity of solution) [ 3 ] . When considering conflict situations that are to be resolved by human intervention, urgency and driver response are particularly important. The chosen conflict situations implemented in our toolkit are listed in Table I (without values of criticality and predictability). No. Name Urgency Response 2 Sensor failure (total) 3 1-2 5 Lane change from entrance ramp not possible 3 3 7 Road narrows (detected by on-board sensors) 3 2 9 Danger zone/obstacle ahead 3 1-3 10 Loss of reference signals (e.g. lane markings missing) 3 1 TABLE I : Conflicts with highest ratings for urgency and driver response from Gold et al. [ 3 ] . III Purpose To address the problem of the lack of data for conflict situations in autonomous driving, our intention is to create a synthetic, reproducible, adaptive database covering various setups in a virtual environment. Comprehensive conflict data is vital for enhancing safety measures, ensuring regulatory compliance, understanding system limitations, advancing technology development, building public trust, and establishing comparative safety metrics in the evolving landscape of av . We narrow down the selection of conflicts for the simulation following the classification in Gold et al. [ 3 ] and choose the conflicts to implement based on higher ratings of urgency and driver response, listed in Table I . The simulation of av can be done in different ways [ 27 ] , mostly based on game engines (e.g. Unity, Unreal Engine). A well-known example of a comprehensive simulator is CARLA [ 2 ] , which is open-source and actively maintained, and it offers a wide range of possibilities for hmi ( hmi ) investigations during conflict situations in the future, e.g. with vr ( vr ) integration [ 28 ] . The different nature of the simulated conflicts requires various technical implementations in CARLA. This includes the setup and

Read the complete text