Paper 2605.26114
MobileGym: A Verifiable and Highly Parallel Simulation Platform for Mobile GUI Agent Research
- Published
- May 2026
- Research lab
- Independent
- Citations
- 2
- GitHub
- 750 stars
01 In brief
Summary
MOBILEGYM is a browser-hosted, lightweight simulation platform for mobile GUI agent research, targeting interaction fidelity without replicating proprietary backends.
It enables verifiable outcome signals via deterministic state-based judging over structured JSON state, and scalable online RL through low-cost parallel rollouts.
The environment state is captured, configured, forked, and compared as structured JSON; a single server can host hundreds of parallel instances (~400 MB each, ~3 s cold start).
A layered state model and declarative task framework support programmability, and a single programmatic judging mechanism provides deterministic evaluation verdicts and dense RL rewards.
The MOBILEGYM-BENCH suite includes 416 parameterized task templates (256 test + 160 train) across 28 apps, with deterministic judges and a structured AnswerSheet protocol.
In a Sim-to-Real case study, GRPO on Qwen3-VL-4B-Instruct gained +12.8 points on the 256-task test set, and on a 59-task real-device signal subset, real-device execution retained 95.1% of the simulation-side training gain.
The platform addresses limitations of emulator-based and real-device environments by making everyday-app state readable, writable, forkable, and consequence-free.
02 From the paper
Abstract
We present MobileGym, a browser-hosted, lightweight, fully controllable environment for everyday mobile use, targeting interaction fidelity without replicating proprietary backends. It enables two capabilities previously out of reach for everyday apps: verifiable outcome signals through deterministic state-based judging over structured JSON state, and scalable online RL through low-cost parallel rollouts. The full environment state is captured, configured, forked, and compared as structured JSON, and a single server can host hundreds of parallel instances, with about 400 MB memory per instance and about 3 s cold start. A layered state model and a declarative task-definition framework keep state programmability and task creation practical at scale, and a single programmatic judging mechanism delivers both deterministic evaluation verdicts and dense RL rewards. The accompanying MobileGym-Bench provides 416 parameterized task templates, including 256 test and 160 train templates, over 28 apps, with deterministic judges and a structured AnswerSheet protocol that avoids free-text matching failures. In a Sim-to-Real case study, GRPO on Qwen3-VL-4B-Instruct gains +12.8 percentage points on the 256-task test set, and on a 59-task real-device signal subset, real-device execution retains 95.1% of the simulation-side training gain. Project page: https://mobilegym.github.io.