PhoneWorld: From Real-App Trajectories to Dynamic and Verifiable Environments for Phone-Use Agents
Summary
The paper introduces PhoneWorld, a trace-grounded framework that converts real-app interaction trajectories into runnable, resettable, and verifiable Android environments spanning 34 apps across 16 domains, enabling benchmark evaluation, verified trajectory generation, and online RL for phone-use agents.
View Cached Full Text
Cached at: 11/20/26, 09:12 AM
# PhoneWorld: From Real-App Trajectories to Dynamic and Verifiable Environments for Phone-Use Agents
Source: [https://arxiv.org/html/2605.29486](https://arxiv.org/html/2605.29486)
Yuxuan Liu∗Affiliation:Gaoling School of Artificial Intelligence, Renmin University of ChinaAffiliation:Tencent HunyuanJunyi Li∗Affiliation:Tencent HunyuanPengyuan Lyu∗Affiliation:Tencent HunyuanChenxin Li∗Affiliation:Tencent HunyuanYiduo GuoAffiliation:Tencent HunyuanZhengyao FangAffiliation:Tencent HunyuanYang DingAffiliation:Tencent HunyuanYi ZhangAffiliation:Tencent HunyuanWeinong WangAffiliation:Tencent HunyuanHuawen ShenAffiliation:Tencent HunyuanXingran ZhouAffiliation:Tencent HunyuanLiang WuAffiliation:Tencent HunyuanFei TangAffiliation:Tencent HunyuanSunqi FanAffiliation:Tencent HunyuanShangpin PengAffiliation:Tencent HunyuanZheng RuanAffiliation:Tencent HunyuanAnran ZhangAffiliation:Tencent HunyuanChengquan Zhang†Affiliation:Tencent HunyuanHan HuAffiliation:Tencent HunyuanBenyou WangAffiliation:The Chinese University of Hong Kong, ShenzhenJi\-Rong WenAffiliation:Gaoling School of Artificial Intelligence, Renmin University of ChinaRui Yan‡Affiliation:Wuhan UniversityZhengyang Tang‡Affiliation:Tencent HunyuanAffiliation:The Chinese University of Hong Kong, Shenzhen
###### Abstract
Real applications provide the training setting closest to phone\-agent deployment, but are difficult to reset, scale safely, and verify programmatically\. Static screenshots and interaction trajectories preserve realistic evidence but cannot generate new experience\. We introducePhoneWorld, a trace\-grounded framework that converts such evidence into runnable, resettable, and verifiable Android environments\. PhoneWorld induces a usage\-weighted interaction skeleton from observed pages, transitions, and state\-changing operations; translates it into a behavior\-grounded app specification; realizes the specification through an autonomous build–inspect–repair loop; and synthesizes executable tasks with programmatic verifiers\. The resulting suite spans 34 consumer\-facing apps across 16 domains and supports an audited online benchmark, verified trajectory generation, and online RL through common reset and verification interfaces\. Evaluations with diverse general and open\-source GUI agents show that PhoneWorld supports reliable end\-to\-end online interaction and exposes capabilities complementary to AndroidWorld\. Controlled SFT experiments further show that PhoneWorld trajectories complement AndroidWorld supervision, transfer across online and offline benchmarks, and become more effective as data volume and app coverage increase\. Under a matched RL budget, combining PhoneWorld mock\-app rollouts with real\-app rollouts improves performance over real\-app RL alone on both real\-phone tasks and AndroidWorld\. Together, these results demonstrate that trace\-grounded executable abstraction can bridge realistic mobile behavior and scalable agent learning, turning limited real\-app evidence into a growing supply of controllable and verifiable environments for training and evaluation\.
††∗\*: Equal contribution\.†††\{\\dagger\}: Project Lead\.††‡\{\\ddagger\}: Corresponding authors\.## 1Introduction
Mobile agents have advanced rapidly from single\-screen perception—such as screen question answering, UI\-element grounding, and isolated clicking—to autonomous systems that can execute complex, multi\-step tasks with little human intervention\([Wang et al\., 2024](https://arxiv.org/html/2605.29486#bib.bib13);[Wang et al\., 2025c](https://arxiv.org/html/2605.29486#bib.bib14);[Zhang et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib12);[Qin et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib10);[Wang et al\., 2025a](https://arxiv.org/html/2605.29486#bib.bib11);[Zhou et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib15);[Xu et al\., 2026a](https://arxiv.org/html/2605.29486#bib.bib18);[Liu et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib38)\)\. Learning these capabilities depends critically on interaction data, whose scale and diversity are determined by the environments in which agents act\. Because deployment ultimately occurs in real applications, improving real\-world assistance requires agents to learn authentic interfaces, content, navigation structures, and state transitions; direct real\-app interaction therefore provides the closest training setting to deployment\. Alongside this shift in capability, mobile\-agent research has expanded from single\-screen understanding to learning from static offline trajectories, and more recently to dynamic data collection and online training in executable mobile environments\([Xu et al\., 2025b](https://arxiv.org/html/2605.29486#bib.bib22);[Shi et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib23);[Luo et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib27);[Chen et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib24);[Xi et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib28)\)\.
The shift from static supervision to dynamic interaction, however, exposes a fundamental tension: the environments closest to real\-world deployment are also the most difficult to use for training at scale\. Concretely, production apps are difficult to reset and run safely in parallel, and rarely expose programmatic task verification, limiting scalable data collection and online training\. AndroidWorld\([Rawles et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib1)\)and MobileWorld\([Kong et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib2)\)enable reproducible online evaluation through reset and programmatic verification, but rely primarily on relatively simple open\-source or system apps, leaving a substantial gap from widely deployed consumer apps in functional scope, page structure, and interaction logic\. AndroidAgentArena\([Chai et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib36)\)and AndroidDaily\([Sui et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib37)\)instead evaluate live production or closed\-source apps, narrowing the realism gap\. Yet such environments remain difficult to reset and do not readily support programmatic verification\. This leaves three questions: \(1\)Static\-to\-Dynamic Environment Synthesis: Can static real\-app screenshots and trajectories be transformed into runnable mock apps that preserve their task\-relevant pages, navigation paths, operations, and state changes; \(2\)Controllable and Verifiable Environment Support: Can these apps be equipped with resettable state, executable tasks, and programmatic verifiers to support evaluation, verified data collection, and online RL; and \(3\)Transferable Training Utility: Can supervision and rewards produced in these environments transfer beyond the mock apps and complement real\-app\-derived training?
To bridge this realism–controllability gap, we introducePhoneWorld, a trace\-grounded environment synthesis framework that distills the task\-relevant structure and behavior of real applications into dynamic, resettable, and programmatically verifiable mobile environments\. As shown in Figure[1](https://arxiv.org/html/2605.29486#S1.F1), PhoneWorld operationalizes this framework through four stages: \(1\)Usage\-Weighted Skeleton Inductionidentifies behaviorally meaningful page types and recovers their visitation frequencies, navigation graph, and state\-changing operations\. \(2\)Behavior\-Grounded App Specificationtranslates this skeleton into prioritized page PRDs, reusable functional components, and runtime content and state specifications\. \(3\)Agentic Environment Realizationautonomously implements and repairs the specification as a runnable Android app, followed by final human acceptance testing\. \(4\)Verifier\-Coupled Task Synthesisderives executable tasks and programmatic verifiers from the same content and mutable state, enabling verified rollouts for data generation, reinforcement learning, and evaluation\.
The resulting PhoneWorld suite comprises 34 consumer\-facing Android environments spanning 16 domains\. Each environment couples a runnable interface with deterministic content, resettable mutable state, grounded tasks, and programmatic verifiers, enabling the same infrastructure to support evaluation, verified rollout generation, and online RL\. The current suite contains a 120\-task audited benchmark, a pool of 7,936 executable tasks, and 3,354 verifier\-confirmed successful trajectories\. Crucially, these environments provide transferable training benefits\. Under a 36K\-step data budget, replacing 10K steps of AndroidWorld supervision with 10K PhoneWorld rollout steps improves success by 52\.5 points on PhoneWorld and 14\.7 points on AndroidWorld\. Holding the PhoneWorld budget at 10K steps, expanding app coverage from 5 to 34 yields further gains of 18\.3 points on PhoneWorld and 10\.4 points on AndroidWorld, demonstrating that environment diversity contributes beyond data volume\. This benefit also extends to online learning: under the same 50\-step RL budget, a 50/50 real/mock rollout mixture improves a separate 4B agent over real\-app RL alone, from 40\.67% to 45\.33% on a 150\-task real\-phone evaluation and from 77\.2% to 83\.2% on AndroidWorld\. Finally, a four\-app audit shows that the generated apps preserve 97\.7% rendered, 92\.4% navigable, and 73\.0% functional coverage of the observed page\-visit mass\.
Our contributions are threefold:
• We introduce atrace\-grounded executable abstraction frameworkthat transforms static real\-app screenshots and trajectories into dynamic, stateful Android environments through Usage\-Weighted Skeleton Induction, Behavior\-Grounded App Specification, and autonomous Agentic Environment Realization with final human acceptance testing\.
• We constructPhoneWorld, a unified suite of 34 consumer\-facing Android environments across 16 domains\. Unified reset, task, and verifier interfaces jointly supportverified data generation, online reinforcement learning, and task evaluation, including a 120\-task audited benchmark\.
• We demonstrate thatPhoneWorld provides effective and transferable training signals\. Controlled supervised\-training experiments show gains on both PhoneWorld and AndroidWorld under a matched budget, with further improvements from broader app coverage at a fixed data budget\. Under the same total RL budget, incorporating PhoneWorld mock\-app interaction into a mixed real–mock recipe further improves both real\-phone evaluation and AndroidWorld over real\-app RL alone\.
Figure 1:PhoneWorld transforms static real\-app evidence into dynamic online environments\.\(a\) Real Interaction Evidence: representative screenshots and exploratory\-use trajectories collected from a real app\.\(b\) Usage\-Weighted Skeleton Induction: page types, visitation frequencies, navigation edges, and state\-changing operations are induced from the observations\.\(c\) Behavior\-Grounded App Specification: the skeleton is translated into prioritized page, transition, content, and state\-effect contracts together with reusable interaction primitives\.\(d\) Agentic Environment Realization: a coding agent implements and iteratively repairs a runnable Android environment, followed by final human acceptance testing\.\(e\) Verifier\-Coupled Task Synthesis: tasks and programmatic verifiers are synthesized from the same content and mutable state, enabling online evaluation, rollout generation, and RL\.
## 2Related Work
Training capable phone\-use agents requires more than static screenshots or demonstration trajectories: agents need a large supply of interactive environments in which they can execute multi\-step actions, observe state changes, and receive reliable automatic feedback\. This requirement has motivated a shift from static mobile benchmarks and offline trajectory learning\([Deng et al\., 2024](https://arxiv.org/html/2605.29486#bib.bib6);[Xu et al\., 2025a](https://arxiv.org/html/2605.29486#bib.bib7);[Xu et al\., 2025c](https://arxiv.org/html/2605.29486#bib.bib8);[Zhang et al\., 2026a](https://arxiv.org/html/2605.29486#bib.bib33);[Lu et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib34);[Wang et al\., 2025b](https://arxiv.org/html/2605.29486#bib.bib35);[Huang et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib25);[Liu et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib26)\)toward online interaction and reinforcement learning for GUI agents\([Xu et al\., 2025b](https://arxiv.org/html/2605.29486#bib.bib22);[Shi et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib23);[Wang et al\., 2025a](https://arxiv.org/html/2605.29486#bib.bib11);[Gu et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib16);[Venus Team et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib17);[Xu et al\., 2026b](https://arxiv.org/html/2605.29486#bib.bib21);[Tang et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib39)\)\. The central artifact in this setting is therefore no longer only a task set or a dataset, but a verifiable environment that can repeatedly produce new interaction experience\.
Existing mobile\-agent benchmarks have made important progress toward verifiable online environments\. AndroidWorld\([Rawles et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib1)\)and MobileWorld\([Kong et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib2)\)provide resettable Android tasks with programmatic checks, while MobileBench\-OL\([Wu et al\., 2026a](https://arxiv.org/html/2605.29486#bib.bib29)\)emphasizes practical mobile task execution\. Their controllability, however, typically depends on a limited set of simple open\-source, system, or locally instrumented apps\. Conversely, Android Agent Arena\([Chai et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib36)\)and AndroidDaily\([Sui et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib37)\)move closer to live or closed\-source production apps, improving realism but making robust reset, safe parallel execution, and deterministic verification harder\. This exposes a core realism–controllability trade\-off: real apps are most relevant for training but hardest to scale\.
A complementary line of work studies how to generate or synthesize interactive environments for agents at scale\([Zala et al\., 2024](https://arxiv.org/html/2605.29486#bib.bib9);[Wang et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib4);[Cao et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib5);[Xu et al\., 2026c](https://arxiv.org/html/2605.29486#bib.bib20)\)\. InfiniteWeb\([Zhang et al\., 2026b](https://arxiv.org/html/2605.29486#bib.bib30)\)generates functional multi\-page websites with task\-centric specifications and verifiable evaluators, AutoWebWorld\([Wu et al\., 2026b](https://arxiv.org/html/2605.29486#bib.bib31)\)translates finite\-state environment models into interactive websites, GUI Exploration Lab\([Yan et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib32)\)builds controllable simulators for studying screen navigation, and Agent\-World\([Dong et al\., 2026](https://arxiv.org/html/2605.29486#bib.bib3)\)synthesizes tool\- and database\-based environments for general agent training and evaluation\. PhoneWorld brings this idea to phone\-use agents by grounding environment synthesis in real mobile interaction evidence: it abstracts observed pages, navigation paths, operations, and state effects from real\-app trajectories, then realizes them as executable Android environments that support evaluation, verified trajectory generation, and online RL through the same reset and verification interfaces\.
## 3Method: Trace\-Grounded Executable Abstraction
PhoneWorld frames environment construction as*trace\-grounded executable abstraction*: from partial observations of a production app, it preserves task\-relevant pages, navigation, visible semantics, and state effects while adding reset and verification interfaces\. Rather than cloning the full service, PhoneWorld induces a usage\-weighted interaction skeleton, specifies executable behavior, realizes it as a stateful Android environment, and synthesizes tasks and verifiers from the resulting state space\. Figure[1](https://arxiv.org/html/2605.29486#S1.F1)summarizes the four stages, and Figure[3](https://arxiv.org/html/2605.29486#A1.F3)gives a concrete example\.
### 3\.1From Static Traces to Dynamic Environments
For a target app, let𝒮\\mathcal\{S\}denote its representative screenshot corpus and let𝒯=\{τi\}i=1N\\mathcal\{T\}=\\\{\\tau\_\{i\}\\\}\_\{i=1\}^\{N\}denote static real\-device trajectories\. A trajectory isτi=\(gi,oi,0,ai,0,oi,1,…,ai,Li−1,oi,Li\)\\tau\_\{i\}=\(g\_\{i\},o\_\{i,0\},a\_\{i,0\},o\_\{i,1\},\\ldots,a\_\{i,L\_\{i\}\-1\},o\_\{i,L\_\{i\}\}\), wheregig\_\{i\}is the user instruction,oi,to\_\{i,t\}is a screenshot, andai,ta\_\{i,t\}is a touch or system action\. Together,𝒟=\(𝒮,𝒯\)\\mathcal\{D\}=\(\\mathcal\{S\},\\mathcal\{T\}\)provides only a static view of the realistic production app\. It captures rendered states and observed transitions, but does not constitute an executable model of the app’s interaction dynamics—how it responds to new action sequences, evolves persistent state, or supports repeated exploration\.
PhoneWorld first induces aUsage\-Weighted Interaction Skeleton𝒦=\(𝒫,𝒢,𝒵\)\\mathcal\{K\}=\(\\mathcal\{P\},\\mathcal\{G\},\\mathcal\{Z\}\), where𝒫\\mathcal\{P\}is a page\-type inventory,𝒢\\mathcal\{G\}is a weighted navigation graph, and𝒵\\mathcal\{Z\}captures action\-conditioned state effects\.Behavior\-Grounded App Specificationthen translates this skeleton into an executable app specification𝚺\\bm\{\\Sigma\}, which defines the page\-level interface, navigation logic, reusable interaction primitives, content organization, mutable\-state semantics, and reset behavior required by the selected workflows\.Agentic Environment Realizationinstantiates this specification as a runnable and resettable surrogate environmentℰ^\\widehat\{\\mathcal\{E\}\}\. Finally,Verifier\-Coupled Task Synthesisderives executable tasks, programmatic verifiers, and verified rollouts from the resulting environment:
𝒟→induce𝒦→specify𝚺→realizeℰ^→synthesize\(𝒬,𝒱,ℛ\+\)\.\\mathcal\{D\}\\xrightarrow\{\\;\\mathrm\{induce\}\\;\}\\mathcal\{K\}\\xrightarrow\{\\;\\mathrm\{specify\}\\;\}\\bm\{\\Sigma\}\\xrightarrow\{\\;\\mathrm\{realize\}\\;\}\\widehat\{\\mathcal\{E\}\}\\xrightarrow\{\\;\\mathrm\{synthesize\}\\;\}\\bigl\(\\mathcal\{Q\},\\mathcal\{V\},\\mathcal\{R\}^\{\+\}\\bigr\)\.\(1\)Here,𝒬\\mathcal\{Q\}contains executable tasks,𝒱\\mathcal\{V\}their programmatic verifiers, andℛ\+\\mathcal\{R\}^\{\+\}verifier\-confirmed successful trajectories\. The distinction between𝒦\\mathcal\{K\}and𝚺\\bm\{\\Sigma\}is central:𝒦\\mathcal\{K\}determines what task\-relevant behavior should be preserved from the observed evidence, whereas𝚺\\bm\{\\Sigma\}specifies how that behavior should be implemented as an executable app\. We refer toℰ^\\widehat\{\\mathcal\{E\}\}as a behavioral surrogate because it preserves the task\-relevant interaction surface of the production app for controlled learning, rather than attempting to reproduce the complete product\.
### 3\.2Usage\-Weighted Skeleton Induction
The induction stage determines*what should survive abstraction*\. Screenshots reveal the visual organization and available affordances of individual pages, while trajectories reveal the app’s operational structure: how frequently pages are visited, how they are connected within common workflows, and how user actions change the app state\. PhoneWorld combines these signals into an*interaction skeleton*that prioritizes empirically important behavior instead of treating every production feature as equally necessary\.
We first organize representative screenshots into an extensible, app\-specific taxonomy of approximately 25–30 page types, such as profile, search, and detail pages\. A vision\-language model \(VLM\) then maps the screenshot corpus into this taxonomy\. The taxonomy abstracts transient visual variations into behaviorally meaningful page types\. Screens showing different content instances or scroll positions are grouped together when they support the same task\-relevant interactions, while visually similar screens remain separate when their available operations or navigation roles differ\.
Formally, letc\(o\)∈𝒫c\(o\)\\in\\mathcal\{P\}map a screenshot to a page type in the recovered taxonomy\. We estimate normalized page frequency and transition weight as
f^\(p\)\\displaystyle\\widehat\{f\}\(p\)=1M∑i=1N∑t=0Li\[c\(oi,t\)=p\],M=∑i=1N\(Li\+1\),\\displaystyle=\\frac\{1\}\{M\}\\sum\_\{i=1\}^\{N\}\\sum\_\{t=0\}^\{L\_\{i\}\}\\mathbf\{1\}\\\!\\left\[c\(o\_\{i,t\}\)=p\\right\],\\qquad M=\\sum\_\{i=1\}^\{N\}\(L\_\{i\}\+1\),\(2\)w\(p,q\)\\displaystyle w\(p,q\)=∑i=1N∑t=0Li−1\[c\(oi,t\)=p,c\(oi,t\+1\)=q\]\.\\displaystyle=\\sum\_\{i=1\}^\{N\}\\sum\_\{t=0\}^\{L\_\{i\}\-1\}\\mathbf\{1\}\\\!\\left\[c\(o\_\{i,t\}\)=p,\\;c\(o\_\{i,t\+1\}\)=q\\right\]\.\(3\)These statistics define a weighted page graph𝒢=\(𝒫,ℰnav,w\)\\mathcal\{G\}=\(\\mathcal\{P\},\\mathcal\{E\}^\{\\mathrm\{nav\}\},w\), with\(p,q\)∈ℰnav\(p,q\)\\in\\mathcal\{E\}^\{\\mathrm\{nav\}\}wheneverw\(p,q\)\>0w\(p,q\)\>0\. The node frequencies quantify how prominently each page appears in observed use, while the edge weights capture how often page pairs are traversed consecutively\. Together, they recover the dominant pages and navigation paths of the app\.
Page visitation and navigation statistics alone do not capture the semantic effects of operations such as favorite, follow or order placement\. We therefore associate important interactions with state\-effect contractsz=\(p,a,ψz\)∈𝒵z=\(p,a,\\psi\_\{z\}\)\\in\\mathcal\{Z\}, whereppis the source page,aais the user action, andψz\\psi\_\{z\}describes the semantic post\-condition to preserve\. For example, a favorite action should make the target entity appear in the user’s collection, and an add\-to\-cart action should update both cart membership and quantity\. These contracts connect the observed interaction skeleton to the mutable\-state design used later for verification\. A worked construction example appears in Appendix[A\.1](https://arxiv.org/html/2605.29486#A1.SS1)\.
### 3\.3Behavior\-Grounded App Specification
The interaction skeleton identifies what to preserve but not how to implement it\. PhoneWorld therefore converts it into an executable app specification𝚺\\bm\{\\Sigma\}covering three aspects:*how to design app, how to implement functional components, and how to prepare runtime data\.*
We first map the page\-frequency ranking to three construction tiers: P0 pages and their connecting paths form the behavioral backbone, P1 pages extend common workflows, and P2 pages cover selected long\-tail functionality\. For each selected page, a VLM turns representative screenshots into a structured PRD specifying its layout, controls, transitions, content fields, and state effects\. Two condensed examples of the resulting page specifications are provided in Appendix[A\.2](https://arxiv.org/html/2605.29486#A1.SS2)\. Recurring requirements are mapped to an evolving library of reusable UI and logic components, such as search, feeds, comments, messaging, carts, and checkout\. As more apps are constructed, recurring patterns and validated fixes are consolidated into this library, allowing subsequent apps to inherit increasingly reliable components while reducing repeated implementation effort\. Finally, we prepare runtime data by separating immutable app contentCCfrom mutable user statemtm\_\{t\}\. For each app, its PRDs determine the data schema and fields required by the supported pages and workflows\. The immutable store contains task\-relevant entities, attributes, and relations for browsing, detail\-page rendering, and deterministic offline BM25 search without external services\. The SQLite\-backed mutable store records action\-induced updates, such as favorites, comments, messages, cart contents, and orders, and exposes the same state to both the interface and downstream verifiers\. Before each task, reset restores a fixed initial snapshot, providing reproducible initial conditions and repeatable execution\. Together, these elements preserve task\-relevant app design and behavior without reproducing the entire product\.
### 3\.4Agentic Environment Realization
A coding agent translates the app specification into Kotlin and Jetpack Compose through an iterative build–inspect–repair process\. It first scaffolds the Android project and data layer, instantiates applicable components from the reusable library, and implements the P0 pages and their end\-to\-end paths before extending coverage to P1 and selected P2 functionality\. The agent then connects page routes, interface controls, content queries, state updates, and reset logic according to the PRDs\.
After each implementation pass, the agent compiles and installs the APK and runs smoke tests over launch, navigation, search, and representative state\-changing operations\. Compiler messages, runtime failures, broken routes, inactive controls, rendering problems, and inconsistent state updates are detected by the agent and used to guide the next repair pass\. This closed loop allows the agent to diagnose failures, revise the implementation, and iteratively optimize the app until the specified workflows execute correctly, without human intervention during construction\. Recurring failures and validated solutions are incorporated into the construction checklist and reusable component library, so later apps benefit from earlier implementation experience\. The resulting app provides deterministic local content, persistent action effects, and a reset operation that restores a known initial snapshot, allowing the same workflows to be executed repeatedly\. Finally, human acceptance testing checks whether any functional defects remain in the principal real–mock workflows\.
### 3\.5Verifier\-Coupled Task Synthesis
A runnable surrogate becomes a learning environment only when goals, executable paths, and success conditions agree\. PhoneWorld therefore synthesizes each task jointly with its verifier rather than generating an instruction first and attaching an evaluator afterward\. Immutable content determines which entities and answers exist, page specifications constrain the available paths and controls, and the mutable\-state schema exposes outcomes that can be checked\.
A taskq=\(gq,xq0,Hq,vq\)q=\(g\_\{q\},x\_\{q\}^\{0\},H\_\{q\},v\_\{q\}\)consists of a natural\-language goalgqg\_\{q\}, a reset configurationxq0x\_\{q\}^\{0\}, a maximum interaction horizonHqH\_\{q\}, and a programmatic verifiervqv\_\{q\}\. Givenqq, an agent policyπ\\piinteracts with the environmentℰ^\\widehat\{\\mathcal\{E\}\}for at mostHqH\_\{q\}steps and produces a trajectoryξ=\(o0,a0,…,oT,yT,mT\)\\xi=\(o\_\{0\},a\_\{0\},\\ldots,o\_\{T\},y\_\{T\},m\_\{T\}\), whereoto\_\{t\}andata\_\{t\}are the screen observation and action at steptt,yTy\_\{T\}is an optional final answer, andmTm\_\{T\}is the final mutable app state\. The verifier maps this trajectory to a binary task outcome:
\(gq,xq0\)→ℰ^,T≤Hq𝜋ξ→vqVq\(ξ\)=vq\(yT,mT\)∈\{0,1\}\.\(g\_\{q\},x\_\{q\}^\{0\}\)\\xrightarrow\[\\widehat\{\\mathcal\{E\}\},\\,T\\leq H\_\{q\}\]\{\\;\\pi\\;\}\\xi\\xrightarrow\{\\;v\_\{q\}\\;\}V\_\{q\}\(\\xi\)=v\_\{q\}\(y\_\{T\},m\_\{T\}\)\\in\\\{0,1\\\}\.\(4\)This joint construction forms a task–environment–verifier closure: every requested entity is grounded in the app content, every required operation is supported by the app specification, and every success condition is observable in the final answer or environment state\. Answer verifiers test grounded values inyTy\_\{T\}, while database verifiers requiremTm\_\{T\}to contain records satisfying task\-specific predicates\. The resulting binary outcome supports evaluation, filtering successful trajectories for supervised learning, and terminal rewards for online RL\. Appendix[A\.3](https://arxiv.org/html/2605.29486#A1.SS3)provides the complete verifier semantics and construction procedure\.
### 3\.6PhoneWorld Environment Suite
Applying this pipeline and the evolving reusable\-component library, we construct PhoneWorld as a unified suite of 34 mock Android apps spanning 16 consumer\-facing domains\. Rather than serving as standalone APKs, all apps follow a common environment interface that couples deterministic content with resettable mutable state, executable task definitions, and programmatic verifiers\. This common interface preserves app\-specific pages, functions, and data while allowing agents to interact with every environment through the same task\-execution and reset protocol\.
Each environment supports three connected uses\. First, tasks are synthesized from the app’s available content, operations, and state schema, so their completion conditions can be checked programmatically\. Second, agents can execute these tasks at scale, with verifier\-confirmed trajectories retained as training data\. Third, reproducible initialization and programmatic terminal rewards provide a stable environment for online RL\. The current suite contains a 120\-task audited benchmark, a pool of 7,936 generated tasks, and 3,354 verified successful trajectories totaling 36,193 interaction steps, as summarized in Table[5](https://arxiv.org/html/2605.29486#A2.T5)in Appendix[B](https://arxiv.org/html/2605.29486#A2)\.
## 4Experiments
Our experiments follow three main questions\.Q1: can heterogeneous agents interact with and be evaluated by PhoneWorld?Q2: do verified trajectories collected from PhoneWorld improve supervised training?Q3: can the environments themselves provide useful online RL experience on top of real\-app RL?
### 4\.1PhoneWorld supports interactive and verifiable online evaluation
We evaluate seven general agents and five open\-source GUI agents on the complete 120\-task PhoneWorld benchmark and 116\-task AndroidWorld benchmark\. We trainPWAgentby supervised fine\-tuning Qwen3\.5\-9B on a 36K\-step mixture of 26K AndroidWorld and 10K PhoneWorld steps\. Both online evaluations report task success across easy, medium, and hard subsets; an attempt is successful only when its programmatic verifier returns full credit\.
Table 1:Primary online evaluation by task difficulty\.We report strict task success rate \(%\) on both PhoneWorld and AndroidWorld benchmarks\.*Steps*is the mean number of dispatched device actions over the tasks an agent solved on that benchmark\.Table[1](https://arxiv.org/html/2605.29486#S4.T1)demonstrates that PhoneWorld functions as an interactive and programmatically verifiable online benchmark\. All 13 agents can be evaluated end to end across the 120 tasks, validating the integration of environment reset, multi\-step interaction, persistent state updates, and automatic verification\. The broad score range—72\.5–94\.2% for general agents and 36\.6–56\.7% for open\-source agents—shows that the benchmark is discriminative\. All five open\-source agents and PWAgent perform worse on hard than easy tasks; for example, PWAgent decreases from 77\.3% to 51\.4%\. PWAgent reaches 65\.0% on PhoneWorld and 71\.6% on AndroidWorld, exceeding the strongest reported open\-source reference by 8\.3 and 13\.8 points, respectively\.
The*Steps*columns measure execution efficiency on solved tasks\. For 12 of the 13 model rows, successful PhoneWorld tasks require more actions than AndroidWorld tasks, consistent with longer consumer\-app workflows\. Among general agents, stronger PhoneWorld performance also tends to coincide with shorter successful trajectories: Claude Opus 4\.8 averages 11\.9 actions, compared with 15\.9 for DeepSeek\-V4 Flash\. PWAgent averages 11\.4 actions, fewer than GUI\-Owl\-1\.5\-8B \(13\.2\) and UI\-Venus\-Navi\-7B \(15\.6\), so its gain does not come from taking longer successful paths\.
More importantly, relative model rankings change across the two benchmarks\. Among general agents, Claude Opus 4\.8 leads Gemini 3\.5 Flash on PhoneWorld \(94\.2% vs\. 92\.5%\) but trails it on AndroidWorld \(78\.4% vs\. 85\.3%\)\. Among open\-source agents, UI\-Venus\-1\.5\-8B slightly outperforms GUI\-Owl\-1\.5\-8B on PhoneWorld \(56\.7% vs\. 55\.8%\) but trails it on AndroidWorld \(47\.4% vs\. 57\.8%\); UI\-Venus\-Navi\-7B similarly leads GUI\-Owl\-7B on PhoneWorld \(50\.0% vs\. 36\.6%\) but not on AndroidWorld \(44\.8% vs\. 46\.6%\)\. PhoneWorld emphasizes Chinese\-language consumer apps, while AndroidWorld primarily covers English\-language open\-source and system apps\. These rank reversals indicate that the benchmarks expose different agent strengths rather than differing only in overall difficulty\.
### 4\.2PhoneWorld\-generated data improves supervised training
We evaluate PhoneWorld\-generated data through three controlled questions: whether it complements AndroidWorld data, whether performance scales with the number of verified PhoneWorld steps, and whether broader app coverage helps at a fixed PhoneWorld budget\. PhoneWorld and AndroidWorld are online benchmarks, whereas AndroidControl and Internal\-MobileBench \(IMB\) are offline benchmarks\. Together, they cover Chinese\- and English\-language settings as well as both simple benchmark apps and realistic consumer apps\.
#### Mixing PhoneWorld with AndroidWorld\.
Table 2:Matched\-budget data composition\.All settings use 36,193 training steps\. PW and AW report online task success; IMB and AC report offline step success\.Table[2](https://arxiv.org/html/2605.29486#S4.T2)compares three mixtures under the same 36,193\-step total training budget: 36k AndroidWorld steps, 26k AndroidWorld \+ 10k PhoneWorld steps, and 36k PhoneWorld steps\. Replacing 10K of the 36K AndroidWorld steps with PhoneWorld rollouts improves PhoneWorld by 52\.5 points, AndroidWorld by 14\.7 points, Internal\-MobileBench by 17\.7 points, and AndroidControl by 6\.0 points, without increasing the total budget\. Full replacement yields the best PhoneWorld score \(73\.3%\) but lowers AndroidWorld to 46\.6%\. The central result is therefore*complementarity*: PhoneWorld adds transferable interaction diversity, while retaining real\-app\-derived supervision is important for the strongest joint performance\.
Figure 2:Scaling PhoneWorld data and app coverage\.Left: performance as the number of PhoneWorld training steps increases\.Right: performance as the number of source apps increases under a fixed 10K\-step PhoneWorld budget\.
#### Scaling data volume and environment coverage\.
Figure[2](https://arxiv.org/html/2605.29486#S4.F2)separates rollout volume from environment diversity\. Adding 0/10K/20K/36,193 PhoneWorld steps raises PhoneWorld success monotonically from 14\.2 to 64\.2/70\.0/73\.3\. Over the same sweep, Internal\-MobileBench rises from 27\.8 to 33\.2 and AndroidControl from 57\.9 to 59\.3, while AndroidWorld remains near 46%\. More verified experience therefore improves in\-domain competence, yields modest offline\-transfer gains, and does not measurably erode AndroidWorld performance\. With the PhoneWorld budget fixed at 10K steps, expanding coverage from 5 to 34 apps raises PhoneWorld from 46\.7 to 65\.0, Internal\-MobileBench from 14\.9 to 33\.2, and AndroidWorld from 61\.2 to 71\.6, while AndroidControl remains broadly stable at 56\.7–59\.7\. Because data volume is fixed, these gains directly demonstrate the value of environment diversity\. Together, the two sweeps show that more rollouts primarily strengthen in\-domain performance, whereas broader app coverage produces the clearest cross\-benchmark gains\.
### 4\.3Mock\-app RL strengthens real\-app RL
Table 3:Real\-phone and AndroidWorld evaluation\. The upper block reports general\-agent references\. The lower block compares different training recipes\. Single\-App, Cross\-App, and Mini\-App each contain 50 human\-judged tasks\. Avg\. is the unweighted mean of the four reported scores\.The general\-agent block provides reference points on the same evaluations: Gemini 3\.1 Pro has the highest average at 59\.1%, while Seed 2\.0 Pro is strongest on Mini\-App tasks at 60\.0%\. These rows are contextual references rather than controlled comparisons with the three training recipes\.
Under the same 50\-step RL budget, allocating half of the rollouts to PhoneWorld improves the real\-phone average from 40\.67% to 45\.33% and AndroidWorld from 77\.2% to 83\.2% over real\-app RL alone\. Gains are concentrated in single\-app and mini\-app workflows, while cross\-app performance remains weak\. Because the total budget is unchanged, the result shows that verifier\-backed mock\-app RL strengthens, rather than merely enlarges, real\-app RL training\. Appendix Figure[5](https://arxiv.org/html/2605.29486#A4.F5)provides qualitative examples of the resulting improvements in constraint following and cross\-app information transfer\.
### 4\.4Four\-app Real–Mock fidelity case study
Table 4:Real–Mock app fidelityWe assess reconstruction fidelity on Douban, Xiaohongshu, Bilibili, and Meituan Waimai using 963 real trajectories, 6,442 valid page visits, and 94 page types\. RPC, NPC, and FPC measure rendered, navigable, and functional page coverage; DER and TER measure recurring and high\-frequency transition recall; and VF measures visual similarity\. All metrics are weighted by observed usage; Appendix[C](https://arxiv.org/html/2605.29486#A3)provides full definitions\. Aggregated across the four apps, rendered, navigable, and functional coverage reaches 97\.7%, 92\.4%, and 73\.0%, respectively\. High\-frequency routes are preserved better than the broader transition graph \(73\.7% TER versus 59\.7% DER\), with 65\.0% pooled visual fidelity\. PhoneWorld thus provides a*task\-oriented reconstruction*that preserves common pages and workflows as executable interactions while selectively abstracting deeper service\-backed functionality and visual detail\.
## 5Conclusion
PhoneWorld converts real\-app screenshots and trajectories into dynamic, resettable, and verifiable Android environments by preserving task\-relevant pages, navigation, visible semantics, and state effects rather than cloning full services\. The resulting suite serves simultaneously as an online evaluation environment, a verified data\-production environment, and an online\-RL training environment through a common reset and verification interface\. Across fidelity, SFT, and RL studies, PhoneWorld provides training signal beyond the mock apps and is most effective as a scalable complement to real\-app experience, suggesting that progress in phone\-use agents depends on scaling controllable environments as well as data\.
### AI use statement
Generative AI tools were used as research instruments in the PhoneWorld pipeline to assist screenshot taxonomy construction, page\-level specification drafting, Android app implementation, and iterative self\-review\. All generated applications and tasks were checked through compilation, emulator testing, programmatic verification, and final human acceptance testing as described in Section[3](https://arxiv.org/html/2605.29486#S3)\. Generative AI was also used for limited language editing and conversion of the manuscript into the official ICLRLaTeXtemplate\. The authors reviewed all AI\-assisted work and take responsibility for the final text, claims, code, data, and artifacts\.
### Ethics statement
PhoneWorld environments are research mock\-ups rather than production apps\. The planned research release uses distinct package names, synthetic seed data, offline emulator\-only execution, and “Mock” launcher labels where appropriate\. Gated APK access is intended for non\-commercial academic research, and app or brand references identify interaction patterns rather than affiliation or endorsement\.
### Reproducibility statement
Section[3](https://arxiv.org/html/2605.29486#S3)describes environment construction, task synthesis, reset logic, and verification, with worked examples in Appendix[A](https://arxiv.org/html/2605.29486#A1)\. Appendix[B](https://arxiv.org/html/2605.29486#A2)summarizes the evaluation benchmarks and metrics, while Appendix[C](https://arxiv.org/html/2605.29486#A3)defines the fidelity audit\. Release plans and access links for the code, benchmark tasks, runner, documentation, and gated mock\-APK dataset are provided in the Availability section below\.
## Availability
We released the PhoneWorld code, benchmark tasks, runner, and documentation at[https://github\.com/EthanLeo\-LYX/PhoneWorld](https://github.com/EthanLeo-LYX/PhoneWorld)\. The mocked apks are constructed by the first author using personal research resource and will be distributed through a gated Hugging Face dataset at[https://huggingface\.co/datasets/EthanLeoLYX/PhoneWorld\-APKs](https://huggingface.co/datasets/EthanLeoLYX/PhoneWorld-APKs)for non\-commercial academic research\. Each user must accept the dataset terms independently rather than receiving redistributed APK mirrors\. The gated APK release uses distinct package names, synthetic data, offline emulator\-only execution, and “Mock” launcher labels where appropriate to make clear that the apps are research mock environments rather than real consumer apps\.
## References
- Caoet al\.\(2026\)Y\. Cao, D\. Ran, M\. Wu, Y\. Guo, X\. Chen, A\. Li, G\. Cao, G\. Zhi, H\. Yu, L\. Li,et al\.Gui\-genesis: automated synthesis of efficient environments with verifiable rewards for gui agent post\-training\.arXiv preprint arXiv:2602\.14093\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p3.1)\.
- Chaiet al\.\(2026\)Y\. Chai, S\. Tang, H\. Xiao, W\. Lin, H\. Li, J\. Zhang, L\. Liu, P\. Zhao, G\. Liu, G\. Wang, S\. Ren, R\. Han, H\. Zhang, S\. Huang, and H\. LiA3: android agent arena for mobile gui agents with essential\-state procedural evaluation\.InFindings of the Association for Computational Linguistics: ACL 2026,pp\. 3774–3789\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p2.1),[§2](https://arxiv.org/html/2605.29486#S2.p2.1)\.
- Chenet al\.\(2025\)Y\. Chen, Y\. Liu, L\. Zhang, P\. Gao, J\. Luan, and W\. LiuSTEP: success\-rate\-aware trajectory\-efficient policy optimization\.arXiv preprint arXiv:2511\.13091\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1)\.
- Denget al\.\(2024\)S\. Deng, W\. Xu, H\. Sun, W\. Liu, T\. Tan, j\. Liu, A\. Li, J\. Luan, B\. Wang, R\. Yan,et al\.Mobile\-bench: an evaluation benchmark for llm\-based mobile agents\.InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),pp\. 8813–8831\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Donget al\.\(2026\)G\. Dong, J\. Lu, J\. Huang, W\. Zhong, L\. Liu, S\. Huang, Z\. Li, Y\. Zhao, X\. Song, X\. Li, J\. Jin, Y\. Zhu, H\. Wang, F\. Lei, Q\. Luo, M\. Chen, Z\. Chen, J\. Feng, J\. Wen, and Z\. DouAgent\-world: scaling real\-world environment synthesis for evolving general agent intelligence\.arXiv preprint arXiv:2604\.18292\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p3.1)\.
- Guet al\.\(2025\)Z\. Gu, Y\. Ke, Y\. Luo, W\. Li, K\. Liu, K\. Yu, L\. Song, S\. Wu,et al\.UI\-venus technical report: building high\-performance ui agents with rft\.arXiv preprint arXiv:2508\.10833\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Huanget al\.\(2025\)K\. Huang, W\. Xu, Y\. Liu, Q\. Wang, P\. Gao, W\. Liu, J\. Luan, B\. Wang, and B\. AnMobileIPL: enhancing mobile agents thinking process via iterative preference learning\.arXiv preprint arXiv:2505\.12299\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Konget al\.\(2026\)Q\. Kong, X\. Zhang, Z\. Yang, N\. Gao, C\. Liu, P\. Tong, C\. Cai, H\. Zhou, J\. Zhang, L\. Chen, Z\. Liu, S\. Hoi, and Y\. WangMobileWorld: benchmarking autonomous mobile agents in agent\-user interactive and mcp\-augmented environments\.InProceedings of the 64th Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),pp\. 6142–6167\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p2.1),[§2](https://arxiv.org/html/2605.29486#S2.p2.1)\.
- Liet al\.\(2024\)W\. Li, W\. Bishop, A\. Li, C\. Rawles, F\. Campbell\-Ajala, D\. Tyamagundlu, and O\. RivaOn the effects of data scale on ui control agents\.Advances in Neural Information Processing Systems37,pp\. 92130–92154\.Cited by:[§B\.1](https://arxiv.org/html/2605.29486#A2.SS1.p1.1)\.
- Liuet al\.\(2025\)Y\. Liu, H\. Sun, W\. Liu, J\. Luan, B\. Du, and R\. YanMobilesteward: integrating multiple app\-oriented agents with self\-evolution to automate cross\-app instructions\.InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V\. 1,pp\. 883–893\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1)\.
- Liuet al\.\(2026\)Y\. Liu, W\. Xu, K\. Huang, C\. Chen, J\. Zhao, P\. Gao, W\. Liu, J\. Luan, S\. Shang, B\. Du,et al\.Come: empowering channel\-of\-mobile\-experts with informative hybrid\-capabilities reasoning\.arXiv preprint arXiv:2602\.24142\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Luet al\.\(2025\)D\. Lu, Y\. Xu, J\. Wang, H\. Wu, X\. Wang, Z\. Wang, J\. Yang, H\. Su, J\. Chen, J\. Chen,et al\.VideoAgentTrek: computer use pretraining from unlabeled videos\.arXiv preprint arXiv:2510\.19488\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Luoet al\.\(2025\)R\. Luo, L\. Wang, W\. He, L\. Chen, J\. Li, and X\. XiaGui\-r1: a generalist r1\-style vision\-language action model for gui agents\.arXiv preprint arXiv:2504\.10458\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1)\.
- Qinet al\.\(2025\)Y\. Qin, Y\. Ye, J\. Fang, H\. Wang, S\. Liang, S\. Tian, J\. Zhang, J\. Li, Y\. Li, S\. Huang,et al\.Ui\-tars: pioneering automated gui interaction with native agents\.arXiv preprint arXiv:2501\.12326\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1)\.
- Rawleset al\.\(2025\)C\. Rawles, S\. Clinckemaillie, Y\. Chang, J\. Waltz, G\. Lau, M\. Fair, A\. Li, W\. Bishop, W\. Li, F\. Campbell\-Ajala, D\. Toyama, R\. Berry, D\. Tyamagundlu, T\. Lillicrap, and O\. RivaAndroidWorld: a dynamic benchmarking environment for autonomous agents\.InInternational Conference on Learning Representations,Cited by:[§B\.1](https://arxiv.org/html/2605.29486#A2.SS1.p1.1),[§1](https://arxiv.org/html/2605.29486#S1.p2.1),[§2](https://arxiv.org/html/2605.29486#S2.p2.1)\.
- Shiet al\.\(2025\)Y\. Shi, W\. Yu, Z\. Li, Y\. Wang, H\. Zhang, N\. Liu, H\. Mi, and D\. YuMobilegui\-rl: advancing mobile gui agent through reinforcement learning in online environment\.arXiv preprint arXiv:2507\.05720\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1),[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Suiet al\.\(2026\)Y\. Sui, X\. Huang, H\. Li, F\. Xu, J\. Lv, H\. Yan, Y\. Shen, L\. Liu, Z\. Fan, Z\. Meng, J\. Wang, J\. Qi, K\. Tan, Z\. Ge, X\. Zhang, D\. Jiang, and O\. YoshieAndroidDaily: a verifiable benchmark for mobile gui agents on real\-world closed\-source applications\.arXiv preprint arXiv:2605\.27761\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p2.1),[§2](https://arxiv.org/html/2605.29486#S2.p2.1)\.
- Tanget al\.\(2026\)Z\. Tang, X\. Lai, P\. Lyu, X\. Wang, T\. Bai, C\. Li, Y\. Guo, H\. Shen, Y\. Liu, J\. Li,et al\.Training open models for agentic phone use\.arXiv preprint arXiv:2606\.23049\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Venus Teamet al\.\(2026\)Venus Team, C\. Gao, Z\. Gu, Y\. Liu, X\. Qiu, S\. Shen, Y\. Wen, T\. Xia, Z\. Xu, Z\. Zeng,et al\.UI\-venus\-1\.5 technical report\.arXiv preprint arXiv:2602\.09082\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Wanget al\.\(2025a\)H\. Wang, H\. Zou, H\. Song, J\. Feng, J\. Fang, J\. Lu, L\. Liu, Q\. Luo, S\. Liang, S\. Huang,et al\.Ui\-tars\-2 technical report: advancing gui agent with multi\-turn reinforcement learning\.arXiv preprint arXiv:2509\.02544\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1),[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Wanget al\.\(2024\)J\. Wang, H\. Xu, J\. Ye, M\. Yan, W\. Shen, J\. Zhang, F\. Huang, and J\. SangMobile\-agent: autonomous multi\-modal mobile device agent with visual perception\.arXiv preprint arXiv:2401\.16158\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1)\.
- Wanget al\.\(2025b\)X\. Wang, B\. Wang, D\. Lu, J\. Yang, T\. Xie, J\. Wang, J\. Deng, X\. Guo, Y\. Xu, C\. Wu,et al\.Opencua: open foundations for computer\-use agents\.arXiv preprint arXiv:2508\.09123\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Wanget al\.\(2026\)Z\. Wang, C\. Xu, B\. Liu, Y\. Wang, S\. Han, Z\. Yao, H\. Yao, and Y\. HeAgent world model: infinity synthetic environments for agentic reinforcement learning\.arXiv preprint arXiv:2602\.10090\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p3.1)\.
- Wanget al\.\(2025c\)Z\. Wang, H\. Xu, J\. Wang, X\. Zhang, M\. Yan, J\. Zhang, F\. Huang, and H\. JiMobile\-agent\-e: self\-evolving mobile assistant for complex tasks\.arXiv preprint arXiv:2501\.11733\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1)\.
- Wuet al\.\(2026a\)Q\. Wu, Z\. Yang, H\. Li, P\. Gao, W\. Liu, and J\. LuanMobileBench\-ol: a comprehensive chinese benchmark for evaluating mobile gui agents in real\-world environment\.arXiv preprint arXiv:2601\.20335\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p2.1)\.
- Wuet al\.\(2026b\)Y\. Wu, Y\. Peng, Y\. Chen, J\. Ruan, Z\. Zhuang, C\. Yang, J\. Zhang, M\. Chen, Y\. Tseng, Z\. Yu,et al\.Autowebworld: synthesizing infinite verifiable web environments via finite state machines\.arXiv preprint arXiv:2602\.14296\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p3.1)\.
- Xiet al\.\(2026\)Z\. Xi, S\. Liang, Q\. Liu, J\. Zhang, L\. Peng, F\. Nan, M\. Nayim, T\. Zhang, R\. Mundada, L\. Qin,et al\.ToolGym: an open\-world tool\-using environment for scalable agent testing and data curation\.arXiv preprint arXiv:2601\.06328\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1)\.
- Xuet al\.\(2026a\)H\. Xu, X\. Zhang, H\. Liu, J\. Wang, Z\. Zhu, S\. Zhou, X\. Hu, F\. Gao, J\. Cao, Z\. Wang,et al\.Mobile\-agent\-v3\.5: multi\-platform fundamental gui agents\.arXiv preprint arXiv:2602\.16855\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1)\.
- Xuet al\.\(2026b\)W\. Xu, Y\. Feng, H\. Lei, K\. Huang, Y\. Liu, K\. Zhao, X\. Hu, S\. Shang, and B\. AnAppDeltaWorld: transition\-grounded delta code world model for mobile gui agents\.arXiv preprint arXiv:2608\.05891\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Xuet al\.\(2026c\)W\. Xu, K\. Huang, Y\. Feng, J\. Li, Y\. Chen, Y\. Liu, Z\. Jiang, H\. Qu, P\. Gao, W\. Liu,et al\.How mobile world model guides gui agents?\.arXiv preprint arXiv:2605\.10347\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p3.1)\.
- Xuet al\.\(2025a\)W\. Xu, Z\. Jiang, Y\. Liu, P\. Gao, W\. Liu, J\. Luan, Y\. Li, Y\. Liu, B\. Wang, and B\. AnMobile\-bench\-v2: a more realistic and comprehensive benchmark for vlm\-based mobile agents\.arXiv preprint arXiv:2505\.11891\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Xuet al\.\(2025b\)Y\. Xu, X\. Liu, X\. Liu, J\. Fu, H\. Zhang, B\. Jing, S\. Zhang, Y\. Wang, W\. Zhao, and Y\. DongMobilerl: online agentic reinforcement learning for mobile gui agents\.arXiv preprint arXiv:2509\.18119\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1),[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Xuet al\.\(2025c\)Y\. Xu, X\. Liu, X\. Sun, S\. Cheng, H\. Yu, H\. Lai, S\. Zhang, D\. Zhang, J\. Tang, and Y\. DongAndroidlab: training and systematic benchmarking of android autonomous agents\.InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),pp\. 2144–2166\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Yanet al\.\(2025\)H\. Yan, Y\. Shen, X\. Huang, J\. Wang, K\. Tan, Z\. Liang, H\. Li, Z\. Ge, O\. Yoshie, S\. Li,et al\.GUI exploration lab: enhancing screen navigation in agents via multi\-turn reinforcement learning\.arXiv preprint arXiv:2512\.02423\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p3.1)\.
- Zalaet al\.\(2024\)A\. Zala, J\. Cho, H\. Lin, J\. Yoon, and M\. BansalEnvgen: generating and adapting environments via llms for training embodied agents\.arXiv preprint arXiv:2403\.12014\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p3.1)\.
- Zhanget al\.\(2026a\)B\. Zhang, Z\. Shang, Z\. Gao, W\. Zhang, R\. Xie, X\. Ma, T\. Yuan, X\. Wu, S\. Zhu, and Q\. LiTongui: internet\-scale trajectories from multimodal web tutorials for generalized gui agents\.InProceedings of the AAAI Conference on Artificial Intelligence,Vol\.40,pp\. 12367–12375\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p1.1)\.
- Zhanget al\.\(2025\)C\. Zhang, Z\. Yang, J\. Liu, Y\. Li, Y\. Han, X\. Chen, Z\. Huang, B\. Fu, and G\. YuAppagent: multimodal agents as smartphone users\.InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems,pp\. 1–20\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1)\.
- Zhanget al\.\(2026b\)Z\. Zhang, Z\. Wang, X\. Zhang, Z\. Guo, J\. Li, B\. Li, and Y\. LuInfiniteWeb: scalable web environment synthesis for gui agent training\.arXiv preprint arXiv:2601\.04126\.Cited by:[§2](https://arxiv.org/html/2605.29486#S2.p3.1)\.
- Zhouet al\.\(2025\)H\. Zhou, X\. Zhang, P\. Tong, J\. Zhang, L\. Chen, Q\. Kong, C\. Cai, C\. Liu, Y\. Wang, J\. Zhou,et al\.MAI\-ui technical report: real\-world centric foundation gui agents\.arXiv preprint arXiv:2512\.22047\.Cited by:[§1](https://arxiv.org/html/2605.29486#S1.p1.1)\.
## Appendix AAdditional Construction Details
This appendix supplements Section[3](https://arxiv.org/html/2605.29486#S3)with a worked reconstruction example, generated page specifications, and additional task\-verifier details\.
### A\.1Worked Reconstruction Example
Figure[3](https://arxiv.org/html/2605.29486#A1.F3)grounds the complete construction pipeline in a QQ\-like environment\. Starting from representative screenshots and exploratory trajectories, PhoneWorld identifies behaviorally meaningful page types, aggregates page visitation and transition statistics, and assigns P0/P1/P2 construction priorities\. The resulting skeleton guides page\-level PRDs and mock page implementation; the same executable content and mutable state then ground a representative task and its SQLite verifier\. The example illustrates how observed usage determines both what is implemented and what can subsequently be executed and verified\.
Figure 3:Worked example of constructing a QQ\-like PhoneWorld environment\.\(a\) High\-frequency real\-app pages: representative screenshots of the most visited page types identified from deduplicated app screenshots\.\(b\) Page frequency and build priorities: trajectory statistics determine page priorities \(P0/P1/P2\), and each page receives a structured PRD\.\(c\) Transition graph: dominant navigation flows are extracted from sequential page visits in the trajectories\.\(d\) Generated mock pages: corresponding screens are implemented from the PRD and reusable component library\.\(e\) Task and verifier: a representative task requiring information seeking and messaging, checked automatically with a SQLite query\.
### A\.2Generated App Specification Examples
The following two examples are condensed from the automatically generated Douban PRD\. We select the high\-frequency home feed and movie detail pages because together they illustrate both a navigation\-heavy top\-level surface and a stateful, information\-dense entity page\. The examples retain the complete four\-part structure supplied to the construction agent, while condensing repetitive low\-level prose for readability\.
douban/home\_feed\.prdP0419 visits⋅\\cdot3 reference views01layoutheader:drawer, central search, audiobook shortcut, and inbox with unread badgenavigation:Following/Recommended tabs, dismissible banner, five fixed bottom tabs, and floating compose actioncontent:vertical mixed\-card feed for group posts, user updates, ratings, and discussions02interaction\_and\_stateactions:search, messages, tabs, cards, identities, join/follow, expand, social actions, and composeeffects:join/follow persists membership; dismiss removes the banner; publish inserts a user\-visible feed item03transitionssearch:search\_inputinbox:messagesfeed cards:group\_post,group\_detail,user\_profilebottom tabs:home\_media,home\_groups,home\_profilecompose:publish04visualpalette:\#FFFFFF,\#2D9144,\#42BD56,\#333333,\#999999type:titles 16–18 sp; body 14–15 sp; metadata 12 spgeometry:16 dp side gutters; 36 dp avatars; 28 dp bottom\-navigation icons
douban/movie\_detail\.prdP0285 visits⋅\\cdot3 reference views01layouthero:back/overflow controls, poster, localized/original titles, year, genres, country, duration, and ranking badgesstate panel:Want/Watched controls, aggregate rating, histogram, and rating countsdetail stream:playback or ticket sources, synopsis, cast/crew, rankings, and overview/review/discussion tabs02interaction\_and\_stateactions:back, share, ranking, status controls, rating panel, sources, synopsis, cast cards, filters, and reviewseffects:Want/Watched persists in the media library; rating persists both stars and the required short review03transitionsrate:rating\_popuprating panel:comments\_listcast card:person\_detailrank/share:ranking\_list,share\_popupback:originating search, ranking, or media\-list context04visualpalette:poster\-derived hero; white/light\-gray foreground; warm orange rating accentstype:title 20–22 sp; body 14–16 sp; metadata 11–12 spgeometry:12–16 dp section rhythm; poster approximately100×140100\\times 140dp
### A\.3Verifier\-Coupled Task Synthesis
PhoneWorld synthesizes tasks jointly from three construction artifacts: read\-only content determines which entities and answers exist, page PRDs determine which interaction paths are available, and the mutable\-state schema determines which outcomes can be verified\. The generator produces a natural\-language goal, difficulty level, step limit, reset configuration, and verification rule, ensuring that every requested entity exists and every required operation is supported by the app\.
Example of Synthesized Task and Verification Rule\{
"app": "mqq",
"difficulty": "easy",
"max\_steps": 10,
"goal": "In QQ, search for \\"Project Group\\" and tap Favorite\.",
"verification": \{
"type": "sqlite",
"database\_path": "/data/data/com\.phoneuse\.mqq/databases/qq\.db",
"table": "user\_collections",
"rules": \[
\{ "field": "target\_type", "operator": "==", "value": "group" \},
\{ "field": "target\_id", "operator": "==", "value": 4 \}
\]
\}
\}Figure 4:Example of a synthesized verifiable task\. Each task is represented as a structured JSON object\. The verification field specifies a database\-level rule that enables automatic task\-completion checking without manual annotation\.PhoneWorld uses two deterministic verifier types\. Information\-seeking tasks check whether the final response contains values grounded in the read\-only content; state\-changing tasks query SQLite for records satisfying task\-specific predicates\. Cross\-app tasks are synthesized from entities shared by two apps and require retrieving information in one environment before acting in another\. Because every task starts from a known state and admits an automatic outcome check, the same task interface supports benchmark evaluation and verifier\-filtered trajectory collection\.
#### From verifiers to RL rewards\.
Each verifier also defines a binary terminal reward\. Answer\-based tasks score the final response, while state\-changing tasks score the final database state\. Resettable initialization and programmatic rewards make PhoneWorld directly usable for repeatable online RL without manual or model\-based judging\.
## Appendix BAdditional Experimental Details
This appendix summarizes the PhoneWorld environment suite \(Table[5](https://arxiv.org/html/2605.29486#A2.T5)\) and the evaluation settings and metrics used in Section 4\.
Table 5:Summary of the PhoneWorld environment suite\.### B\.1Benchmarks and Metrics
Table[6](https://arxiv.org/html/2605.29486#A2.T6)summarizes the four evaluations\. PhoneWorld and AndroidWorld\([Rawles et al\., 2025](https://arxiv.org/html/2605.29486#bib.bib1)\)are the primary online benchmarks; Internal\-MobileBench and AndroidControl\([Li et al\., 2024](https://arxiv.org/html/2605.29486#bib.bib19)\)are offline transfer diagnostics\.
Table 6:Evaluation benchmarks used in this paper\.PhoneWorld contains 44 easy, 41 medium, and 35 hard tasks; the evaluated AndroidWorld split contains 61 easy, 36 medium, and 19 hard task types\. Online success requires full programmatic credit; partial scores, execution errors, and failed checks count as unsuccessful\. The*Steps*columns in Table[1](https://arxiv.org/html/2605.29486#S4.T1)report the mean number of dispatched device actions over solved tasks, excluding screenshot and shell calls\. Because this average is conditioned on success, it measures execution efficiency only within the subset a model solves\.
## Appendix CReal–Mock Fidelity Metrics
We quantify reconstruction fidelity using visit\-weighted page coverage, transition recall, and visual fidelity\. All metrics are first computed per app\. Aggregate scores pool the corresponding numerators and denominators across apps rather than averaging per\-app percentages\.
#### Page coverage\.
For appaa, let𝒫a\\mathcal\{P\}\_\{a\}be its observed real page types,na\(p\)n\_\{a\}\(p\)the number of real visits to page typepp, andℓa\(p\)∈\{0,1,2,3\}\\ell\_\{a\}\(p\)\\in\\\{0,1,2,3\\\}its manually audited mock implementation level\. L0 denotes an absent page; L1 a rendered proxy without a task\-equivalent reachable page; L2 a reachable and useful page with important semantics or actions missing; and L3 a page whose core task\-relevant interaction produces the expected state change\. For thresholdkk, visit\-weighted page coverage is
PCa\(k\)=100∑p∈𝒫ana\(p\)1\[ℓa\(p\)≥k\]∑p∈𝒫ana\(p\)\.\\operatorname\{PC\}\_\{a\}\(k\)=100\\,\\frac\{\\sum\_\{p\\in\\mathcal\{P\}\_\{a\}\}n\_\{a\}\(p\)\\,\\mathbf\{1\}\[\\ell\_\{a\}\(p\)\\geq k\]\}\{\\sum\_\{p\\in\\mathcal\{P\}\_\{a\}\}n\_\{a\}\(p\)\}\.\(5\)We report RPC=PCa\(1\)=\\operatorname\{PC\}\_\{a\}\(1\), NPC=PCa\(2\)=\\operatorname\{PC\}\_\{a\}\(2\), and FPC=PCa\(3\)=\\operatorname\{PC\}\_\{a\}\(3\)\. Thus, RPC measures visible page presence, NPC adds reachability and task usability, and FPC further requires the core semantic state effect\.
#### Transition recall\.
Letℰa\\mathcal\{E\}\_\{a\}be the directed page transitions observed in real trajectories,na\(e\)n\_\{a\}\(e\)the frequency of transitionee, andsa\(e\)∈\{0,1\}s\_\{a\}\(e\)\\in\\\{0,1\\\}indicate whether the mapped transition is directly supported by the mock according to a static source audit\. At a minimum real\-transition countcc, frequency\-weighted edge recall is
ERa\(c\)=100∑e∈ℰa:na\(e\)≥cna\(e\)sa\(e\)∑e∈ℰa:na\(e\)≥cna\(e\)\.\\operatorname\{ER\}\_\{a\}\(c\)=100\\,\\frac\{\\sum\_\{e\\in\\mathcal\{E\}\_\{a\}:\\,n\_\{a\}\(e\)\\geq c\}n\_\{a\}\(e\)s\_\{a\}\(e\)\}\{\\sum\_\{e\\in\\mathcal\{E\}\_\{a\}:\\,n\_\{a\}\(e\)\\geq c\}n\_\{a\}\(e\)\}\.\(6\)DER isERa\(2\)\\operatorname\{ER\}\_\{a\}\(2\), which evaluates the broader recurring transition graph, and TER isERa\(10\)\\operatorname\{ER\}\_\{a\}\(10\), which isolates the most frequently traversed navigation backbone\. Weighting byna\(e\)n\_\{a\}\(e\)makes a missed common route count more than a missed rare route\.
#### Visual fidelity\.
For each paired real–mock page type, we assignza,p,d∈\{0,1,2\}z\_\{a,p,d\}\\in\\\{0,1,2\\\}on five dimensionsdd: layout, key elements, affordance, content semantics, and visual hierarchy\. Scores 0, 1, and 2 respectively indicate a substantial mismatch, a partial match with noticeable differences, and a close task\-relevant match\. Letℳa⊆𝒫a\\mathcal\{M\}\_\{a\}\\subseteq\\mathcal\{P\}\_\{a\}denote the paired page types andD=5D=5\. Visit\-weighted visual fidelity is
VFa=100∑p∈ℳana\(p\)∑d=1Dza,p,d2D∑p∈ℳana\(p\)\.\\operatorname\{VF\}\_\{a\}=100\\,\\frac\{\\sum\_\{p\\in\\mathcal\{M\}\_\{a\}\}n\_\{a\}\(p\)\\sum\_\{d=1\}^\{D\}z\_\{a,p,d\}\}\{2D\\sum\_\{p\\in\\mathcal\{M\}\_\{a\}\}n\_\{a\}\(p\)\}\.\(7\)
## Appendix DQualitative Analysis of Real\+Mock RL
Figure[5](https://arxiv.org/html/2605.29486#A4.F5)presents two representative real\-phone trajectories that illustrate how Real\+Mock RL changes execution behavior beyond the aggregate success rates reported in Section[4\.3](https://arxiv.org/html/2605.29486#S4.SS3)\. The examples compare the SFT policy with the policy trained by combining real\-app and PhoneWorld mock\-app rollouts under the same total RL budget\.
Figure 5:Qualitative comparison between SFT and Real\+Mock RL on real\-phone tasks\.Top: Real\+Mock RL applies the requested hotel\-price constraint, whereas SFT stops at an unfiltered result page\.Bottom: Real\+Mock RL correctly transfers generated text from Yuanbao to Tencent Docs, whereas SFT inserts stale clipboard content\.#### Constraint following\.
The first task asks the agent to search for budget\-friendly hotels near Shanghai Disneyland in the Tongcheng Travel WeChat mini\-app\. The SFT policy reaches a plausible hotel\-results page but does not enforce the requested price constraint\. The Real\+Mock RL policy instead continues to the filtering interface and sets the maximum hotel price to 150 yuan\. This case illustrates improved retention of task constraints across a multi\-step workflow and more complete use of controls that are not available on the initial search page\.
#### Cross\-app information transfer\.
The second task requires the agent to generate a leave\-request note with Yuanbao and save it in Tencent Docs\. The SFT policy reaches the document editor but fails to copy the newly generated note, causing stale clipboard content to be inserted\. The Real\+Mock RL policy correctly selects and copies the generated note, switches to Tencent Docs, creates a document, and pastes the intended content\. The improvement therefore involves not only navigation across apps, but also maintaining and transferring an intermediate information artifact through the clipboard\.
Together, these examples suggest that mixed\-environment RL does more than encourage broader exploration: it provides useful supervision for satisfying explicit task constraints and executing stateful operations required for cross\-app information transfer\.Similar Articles
PhoneWorld: Scaling Phone-Use Agent Environments
PhoneWorld is a pipeline that transforms real GUI trajectories into controllable mobile environments, enabling scalable creation of phone-use benchmarks. It covers 34 apps across 16 domains and shows that using its supervision improves performance on multiple evaluation benchmarks.
AndroidReality: How Far Are Mobile Agents from the Real World?
Introduces AndroidReality, a perturbation-based framework for evaluating and improving the robustness of mobile agents, with a taxonomy of real-world interface perturbations and a training-free Test-Time Introspective Recovery (TTIR) mechanism.
RecreationWorld: Scalable and Verifiable Environments for Hybrid Computer-Use Agents
RecreationWorld introduces a scalable and verifiable framework for hybrid computer-use agents, providing environments across five platforms and a benchmark for evaluation.
Training Open Models for Agentic Phone Use
PhoneBuddy combines real and mock app environments to train open models for agentic phone use, achieving 45.33% task success rate on real phones through mixed reinforcement learning, showing that mock-app training complements real-app training.
iOSWorld: A Benchmark for Personally Intelligent Phone Agents
Introduces iOSWorld, an interactive native iOS simulator benchmark with persistent user identity across 26 apps, designed to evaluate personalized mobile agent capabilities through 133 tasks of increasing difficulty.