AndroidEnv: A Reinforcement Learning Platform for Android

Daniel Toyama, Philippe Hamel, Anita Gergely, Gheorghe Comanici, Amelia Glaese, Zafarali Ahmed, Tyler Jackson, Shibl Mourad, Doina Precup

🏛 Institutions: Google DeepMind
📅 Date: May 27, 2021
📑 Publisher: arXiv
💻 Env: Mobile
🔑 Keywords: environment reinforcement learning AndroidEnv touchscreen interface Android platform

TLDR

AndroidEnv is an open-source research platform that exposes Android apps through a realistic touchscreen interaction loop for reinforcement learning. The paper frames Android device use as a unified environment family rather than a single fixed benchmark task.

Open paper arXiv Edit on GitHub Report issue