Pattern-Based Reinforcement Learning for Wearable AI Vision Assist Devices on Edge Platforms

This paper presents an AI-powered wearable device designed to assist visually impaired individuals by integrating natural language processing (NLP), computer vision (CV), and reinforcement learning (RL). Built on a Raspberry Pi platform, the system combines object detection (YOLO), facial recognition (DeepFace), and LLM-based query reasoning to provide real-time audio feedback. The proposed solution enhances navigation, awareness, and safety through intelligent perception and context understanding. We detail the system architecture, optimization techniques for low-power execution, and demonstrate its practical effectiveness in real-world scenarios.