Research papers · Posted 2026.05.17 · Preprint
Generating Realistic Safety-Critical Scenarios for Vehicle-Pedestrian Interactions
Generating Realistic Safety-Critical Scenarios for Vehicle-Pedestrian Interactions
Overview
A three-stage approach combines real interaction data, multi-agent pretraining and CARLA reinforcement learning to generate vehicle-pedestrian scenarios. The resulting VPSCI dataset is evaluated for realistic evasive behavior and conflict severity.
Author-supplied source information
Author-supplied information on arXiv.
49 pages, 13 figures, 11 table
Version: v2