Radio-strike: A reinforcement learning game for MIMO beam selection in unreal engine 3-D environments

* Register (or log in) to the AI4G Neural Network to add this session to your agenda or watch the replay
Gaming and other industries are driving the development of sophisticated tools to create virtual worlds, composed of 3-D models, physics engines and other components. This talk promotes the vision that 5G and beyond will benefit from the availability of virtual worlds to leverage machine learning / artificial intelligence (ML/AI) applied to communication networks. As a concrete example of a CAVIAR (“Communication networks and Artificial intelligence immersed in VIrtual or Augmented Reality”) framework, this talk will present “Radio-Strike”, developed with Epic Games’ Unreal Engine and Microsoft’s AirSim simulator. Radio-Strike is used in the UFPA Problem Statement for the 2021 ITU AI/ML in 5G Challenge, in which participants are invited to develop a deep reinforcement learning (RL) agent that plays the role of a base station serving drones, vehicles and pedestrians via MIMO beamforming in a 3-D scenario. This problem targets those interested in designing ML/AI systems that learn from experience and outperform conventional solutions relying only on signal processing. I will provide an introduction to RL applied to MIMO beamforming, explain Radio-Strike and discuss challenges to deploy RL in 5G / 6G networks.