Page 54 - 2024S

UEC Int’l Mini-Conference No.52 47

Achieving Robust Autonomous Driving Through Multi-View Deep

Reinforcement Learning

Kittithammo KRITSANAPHONG , Celimuge WU
∗
Department of Computer and Network Engineering
The University of Electro-Communications
Tokyo, Japan

Keywords: Machine learning, Reinforcement Learning, Autonomous Driving, Sensor Fusion.

Abstract
Autonomous driving technology has made significant advancements. However the challenge of en-
suring robust and consistent performance across varying conditions remains. In order to achieve more
advanced autonomous driving, it is necessary to refer to wide-area information and integrate data from
multiple sensors. This research aims to achieve the robustness and reliability of autonomous driving
systems that cannot be achieved with conventional sensor fusion through the application of multi-view
deep reinforcement learning. By integrating multiple sensor technologies such as LiDAR and Vidar,
the research focuses on optimizing the learning performance and cooperative behavior of autonomous
agents, thereby improving their decision-making capabilities in complex driving scenarios. The study
leverages the CARLA simulator to create a realistic testing environment. Expected results include
significant improvements in the accuracy and reliability of autonomous vehicle control. By sharing data
across multiple views and employing advanced learning algorithms, the system is anticipated to achieve
higher accuracy in real-time decision-making.

∗
The author is supported by (MICH) MEXT Scholarship.