Suicidal Pedestrian: Generation of safety-critical scenarios for autonomous vehicles
Loading...
URL
Journal Title
Journal ISSN
Volume Title
Sähkötekniikan korkeakoulu |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
2023-05-15
Department
Major/Subject
Control, Robotics and Autonomous Systems
Mcode
ELEC3025
Degree programme
AEE - Master’s Programme in Automation and Electrical Engineering (TS2013)
Language
en
Pages
53+2
Series
Abstract
Autonomous driving is appealing due to its significant financial potential and positive social impact. However, developing capable autonomous driving algorithms faces the difficulty of reliability testing because some safety-critical traffic scenarios are particularly challenging to acquire. To this end, this thesis proposes a method to design a suicidal pedestrian agent based on the CARLA simulation engine that can automatically generate pedestrian-related traffic scenarios for autonomous vehicle testing. In this method, the pedestrian is formulated as a reinforcement learning agent that spontaneously seeks collisions with the target vehicle and is trained using a continuous model-free learning algorithm with two custom reward functions. Besides, by allowing the pedestrian freely explore the environment with a constrained initial distance to the vehicle, the pedestrian and autonomous car can be placed anywhere, rendering generated scenarios more diverse. Furthermore, four collision-oriented evaluation metrics are also proposed to verify the performance of the designed suicidal pedestrian and the target vehicle under testing. Experiments on two state-of-the-art autonomous driving algorithms demonstrate that this suicidal pedestrian is effective in finding autonomous vehicle decision errors when cars are exposed to such pedestrian-related traffic scenarios.Description
Supervisor
Ilin, AlexanderThesis advisor
Kujanpää, KallePajarinen, Joni
Keywords
autonomous driving, suicidal pedestrian, traffic scenario generation, reinforcement learning, CARLA simulator