GameLab: AI-Enabled Cloud Gaming Testbed

We present GameLab, an open-source, AI-enabled cloud gaming testbed built on WebRTC. Unlike existing open-source stacks and deployment-oriented pipelines (e.g., GamingAnywhere, Sunshine/-Moonlight, and Unity Render Streaming), GameLab is designed for AI-in-the-loop systems research: it provides programmable interfaces on both the server and client, with well-defined hook points to plug in machine learning modules on demand (e.g., super-resolution, denoising, object detection, QoE estimation, and learned rate control). GameLab also collects detailed transport and application traces for online and offline analyses of gaming sessions and network behavior. To enable reliable objective evaluation in interactive settings, GameLab embeds compact QR-based frame identifiers into the video stream, allowing accurate computation of full-reference quality metrics, such as PSNR, SSIM, and VMAF, even under frame loss, reordering, and duplication. Finally, GameLab supports GPU-based visualization of frames and model outputs for interactive inspection without costly GPU-to-CPU transfers.We demonstrate the versatility of GameLab through multiple end-to-end experiments, including: (i) transparently applying video upscalers to reduce bandwidth, (ii) deploying multiple client-side object detection models, and (iii) implementing a server-side visual-complexity-driven bitrate ladder for rate adaptation.Github: https://github.com/haseebfazal/ai-testbed.git.