AndroidWorld_SR

AndroidWorld SR (Success Rate) is a dynamic benchmarking environment for autonomous Android agents featuring 116 programmatic tasks across 20 real-world Android apps. It evaluates AI agents' mobile device control capabilities through diverse tasks like recording audio, adding expenses, creating map markers, and managing calendars. The benchmark provides durable reward signals and measures success rates for complex multi-step interactions, with baseline agents achieving around 30.6% task completion.