The surge in popularity of online gaming, particularly on smartphones,
underscores the growing trend and demand for immersive gaming
experiences. Online smartphone gaming faces challenges such as poor
Wi-Fi conditions and network handovers, impacting user experiences. In
response, we propose ODIN (On-Device Intelligence), a Neural
Network-driven and QUIC transport layer protocol based gaming proxy to
enrich the users' Quality of Experience (QoE). ODIN integrates a
sophisticated NN-based network quality monitoring framework to predict
Wi-Fi contention accurately, ensuring optimal network selection for
gaming applications and minimizing lags. Live-air experiments
featuring popular Android gaming applications (such as BrawlStars)
confirm ODIN's superiority over legacy smartphones. It surpasses
Multipath TCP (MPTCP) in handling thin-stream applications, showcasing
a seamless zero-touch, zero-lag handover mechanism in the BrawlStars
game.


Evaluation results demonstrate ODIN's consistent provision of superior
gaming QoE while optimizing mobile data usage. Dynamic selection of
the best network interface, guided by NN-based predictions, ensures
seamless handover between Wi-Fi and mobile networks, contributing to
efficient network resource utilization and significant savings in
mobile data consumption. In addition to robust performance, ODIN
introduces ODIN-LITE, a lightweight variant exhibiting a remarkable
25% improvement in power consumption efficiency compared to MPTCP's
full-mesh mode. ODIN enhances gaming Quality of Experience (QoE)
across diverse network conditions.