AROB 2024 paper abstract: The increasing demand for electronic products has led to a surge in end-of-life devices, making efficient recycling crucial for minimizing environmental impact. However, current recycling processes are often tailored to specific models, making adaptation to varying devices complex and costly. This paper addresses the challenge of automating electronic device disassembly using computer vision and action prediction methods. The research explores key components of a robotic disassembly system, including pose estimation, device classification, rotation estimation, gap detection, and action prediction. High accuracy is achieved using segmentation models and supervised learning for known devices, while zero-shot classification and data-driven approaches show promise for handling unseen devices. A large language model (LLM) is introduced for action prediction, demonstrating its ability to adapt to diverse disassembly tasks with 91% accuracy. The results indicate that generalization across device models is possible but varies by method. This study provides a framework for developing flexible and robust robotic systems, paving the way for more sustainable and scalable electronic recycling solutions.