Meta-Reinforcement Learning for UAV-Assisted Mobile Edge Computing of Virtual Reality Services.

Virtual reality (VR) services face significant challenges, such as large-volume data transmission, high computing demands, and extreme latency sensitivity, particularly on wireless VR devices with limited resources. Unmanned aerial vehicle (UAV)-assisted mobile edge computing (MEC) emerges as a promising solution to ease the computational burden on user devices and enable efficient VR service delivery. This paper proposes a UAV-assisted collaborative MEC framework, involving one access point (AP), one UAV, and VR users, catering to various VR scenarios. VR videos are cached in 2D field of view (FoV) format at the AP, while users request 3D FoV rendering. The 3D FoV model is derived via edge computing at the AP, UAV, or VR users. We formulate an optimization problem with the objective of maximizing VR users’ average service completion rate (ASCR) while minimizing UAV energy consumption, subject to UAV movement, energy, and transmit power constraints. Then, we model the problem as a Markov decision process (MDP) and employ meta-reinforcement learning (Meta-RL) for an optimal strategy. Extensive simulations confirm the efficacy of our approach in enhancing the user experience, efficiently leveraging UAV capacities for edge computing and communications. The robust generalization ability of our Meta-RL algorithm ensures suitability for diverse real-world scenarios.