Emergent Pedestrian Safety in a World-Model Driving Agent Under Adversarial Interaction Without Explicit Safety Rewards
Applied Sciences · 17.04.2026 · 16 (8) · 3915
DOI: 10.3390/app16083915 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 17.04.2026. Verfasst von Stefan Zlatinov und 4 weiteren Autor:innen. Die Arbeit nennt 50 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Austrian Federal Ministry for Transport, Innovation and Technology, Austrian Federal Ministry for Digital, Business and Enterprise, Austrian Research Promotion Agency.
Abstract (Originalsprache)
Pedestrian interaction remains a central safety challenge for autonomous driving, particularly under non-compliant or adversarial pedestrian behavior. Existing research and evaluations predominantly test against rule-following pedestrians, leaving a gap in understanding how learning-based agents handle worst-case interactions. We introduce the Jaywalkers Library, a novel configurable benchmark in CARLA with three adversarial pedestrian archetypes (Intruder, Indecisive Crosser, and Protester). We evaluate a DreamerV3 agent trained with sparse rewards, where the only pedestrian-specific signal is a terminal collision penalty. Evaluation employs a frozen-policy protocol with explicit train–test separation. Safety behavior is decomposed into endpoint outcomes, evasion dynamics, and efficiency costs. Under nominal conditions, the agent achieves high route completion and generalizes to an unseen town, whereas under adversarial exposure, an archetype-sensitive evasion strategy emerges. The agent swerves at speed against dynamic pedestrians but decelerates against the slow-moving Protester. Collision rates reveal a counterintuitive difficulty ordering in which the Protester is the hardest, followed by the Intruder, with the Indecisive Crosser as the most survivable. These findings show that a sparse terminal penalty suffices for emergent pedestrian avoidance in a world-model agent, but that effectiveness is bounded by the world model’s ability to predict pedestrian persistence.
Förderung
- Austrian Federal Ministry for Transport, Innovation and Technology
- Austrian Federal Ministry for Digital, Business and Enterprise
- Austrian Research Promotion Agency
- Province of Styria
- Styrian Business Promotion Agency
Zitieren
APA
Zlatinov, S., Nadzinski, G., Latkoska, V. O., Stavrov, D., & Stankovski, M. (2026). Emergent Pedestrian Safety in a World-Model Driving Agent Under Adversarial Interaction Without Explicit Safety Rewards. Applied Sciences, 16(8), 3915. https://doi.org/10.3390/app16083915
Vancouver
Zlatinov S, Nadzinski G, Latkoska VO, Stavrov D, Stankovski M. Emergent Pedestrian Safety in a World-Model Driving Agent Under Adversarial Interaction Without Explicit Safety Rewards. Applied Sciences. 2026;16(8):3915. doi:10.3390/app16083915
BibTeX
@article{Zlatinov2026,
author = {Zlatinov, Stefan and Nadzinski, Gorjan and Latkoska, Vesna Ojleska and Stavrov, Dushko and Stankovski, Mile},
title = {Emergent Pedestrian Safety in a World-Model Driving Agent Under Adversarial Interaction Without Explicit Safety Rewards},
journal = {Applied Sciences},
year = {2026},
volume = {16},
number = {8},
pages = {3915},
doi = {10.3390/app16083915}
}
EUVIMED