[4Yin-B] Poster 7

[4Yin-B-33] Cross-platform Distributed Reinforcement Learning Framework for the Web

〇Tomohiro Hashimoto¹, Yuto Nishizawa¹, Masatoshi Hidaka¹, Takayuki Osa², Tatsuya Harada^1,2 (1. The University of Tokyo, 2. RIKEN)

Keywords:

Machine Learning,Reinforcement Learning,Edge Devices

Reinforcement learning requires immense computational resources, and recent trends have shifted toward utilizing multi-CPU and GPU setups to accelerate environment interactions. Despite the vast aggregate capacity of consumer devices such as PCs and smartphones, these resources remain largely underutilized for data collection. This paper proposes a cross-platform distributed reinforcement learning framework that leverages web browsers to eliminate installation barriers. Designed around the APE-X architecture, the system employs browser-based actors that connect to Unity and MuJoCo environments for seamless data generation. Experiments demonstrate successful reinforcement learning through the parallel execution of multiple edge devices, presenting a novel and scalable use case for consumer hardware in the field.