Automation of Android-based actions using large action models

Despite their ubiquity, mobile devices often rely on limited and rigid interaction methods, hindering accessibility and user experience. Recent advancements in large language models (LLMs) offer a promising solution, enabling natural language interfaces that simplify complex tasks. This research explores the potential of LLMs to automate Android system actions through two innovative approaches: an offline method using a fine-tuned Gemma-2B model and an online approach leveraging Llama-3.3-70B. The proposed work implementation successfully automates various device functions, such as Bluetooth/Wi-Fi toggling and SMS messaging, highlighting the trade-offs between privacy, performance, and network dependency. By addressing the challenges of deploying AI assistants on mobile platforms, this research contributes significantly to enhancing device accessibility and usability, paving the way for more sophisticated AI-driven mobile interactions.