Subverting the Grinding Economy: An Edge-Based Autonomous Deep Reinforcement Learning Agent and Humanized Behavioral Biometrics in Mobile Gaming

Abstract

Modern mobile freemium games increasingly rely on coercive engagement loops and artificially inflated grinding mechanics—a socio-technical phenomenon conceptualized as digital "playbor"—to drive in-app purchases and ad monetization. To protect these revenue streams, commercial game publishers deploy sophisticated behavioral biometric surveillance systems that monitor user touch dynamics to detect and ban automated agents. This paper presents an end-to-end, edge-based autonomous artificial intelligence system that subverts these exploitative game loops while actively evading statistical anti-cheat detection in Cookie Run: Classic. The system combines a sub-millisecond zero-copy X11 display ingestion pipeline (119.0 FPS, 8.4 ms), an invariant screen state machine, and a dual-paradigm deep learning architecture. A lightweight convolutional neural network (CookieRunnerNet, 883k parameters) achieves real-time inference in 1.80 ms on a standard multi-core laptop CPU, achieving 98.0% slide recall and an 84.0% action F1 score. This behavioral cloning backbone subsequently warm-starts a Dueling Deep Q-Network (DuelingCookieQNet) with Prioritized Experience Replay, achieving 86.7% zero-shot policy alignment from frame 1. To bypass automated biometric surveillance, we formulate a stochastic actuation engine incorporating Gaussian contact duration jitter (N(65, 12) ms), biological double-jump refractory limits (> 160 ms), and action hysteresis. Operating without memory injection, root privileges, or server-side telemetry tampering, the agent demonstrates that consumer edge AI can emancipate player time from repetitive playbor, challenge algorithmic biometric profiling, and provide cognitive-assistive capabilities for players with motor impairments.