[B-5A] 無線通信システムA

[B-5-2] Optimized Channel Selection in Vehicular Systems Using Reinforcement Learning Techniques

^○Yun LI, Yuyuan CHANG, Kazuhiko FUKAWA (Tokyo Tech)

この講演は本会「学術奨励賞受賞候補者」の資格対象です。

Keywords:Autonomous driving、wireless communications、V2X、cognitive radio、multi-action reinforcement learning

This study presents a solution to spectrum scarcity in cognitive radio-based V2X communications, which have frequently been compounded by oversimplified V2X models. By utilizing a more precise 3D autonomous driving testbed [1] and a proximal policy optimization (PPO) reinforcement learning (RL) algorithm, we propose and apply a multi-action PPO (MA-PPO) to the complex optimization problems. Computer simulations demonstrate that MA-PPO is superior to the con- ventional multi-action Deep Q-network (MA-DQN) approach in terms of both the stability and data transmission efficiency.