On realistic scenarios for hazard perception of vulnerable road users

Evaluating contextual hazard of vulnerable road users (VRUs) under realistic driving and sensory contexts are critical to the integration of VRU with legacy and automated vehicles. Over the last decades, various synthetic scenarios have been designed and calibrated for microscopic simulators for SUMO mostly focusing on vehicles. Realistic traffic datasets including VRU such as RounD have been used to extract and learn precise driving and hazard patterns but cannot be modified to evaluate the impact of C-ITS safety applications for VRU in the dataset environment. The driving simulator CARLA has been designed to model robotic and sensory context in highly precise driving environment, which makes it perfectly suitable to model VRU in mixed traffic scenarios. However, most of the studies using CARLA focuses primarily on the modeling or the perception of an ego-vehicle (or a VRU) either isolated or under unrealistic traffic. Considering that realistic traffic interacting with VRU is critical to identifying hazard contexts for VRUs, this talk presents an open-source CARLA [1] scenario reproducing the RounD dataset and discuss its benefit to integrate realistic perception of VRUs