Pith. sign in

REVIEW 4 major objections 4 minor 9 cited by

MobiAgent, a 7B/3B decider-grounder pair with a record-replay cache, achieves the best task-completion scores on real Chinese mobile apps, beating far larger general-purpose and specialized GUI models.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

A full-stack mobile agent framework reports state-of-the-art task completion on its own DAG-based benchmark and 2-3x speedups from replaying recorded trajectories.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A credible full-stack mobile agent system with a real SOTA claim, but the claim is measured only on an unreleased, author-curated benchmark, so external validation is needed before the comparison is established. the 4 major comments →

arxiv 2509.00531 v1 pith:VY36LQCY submitted 2025-08-30 cs.MA cs.LG

MobiAgent: A Systematic Framework for Customizable Mobile Agents

classification cs.MA cs.LG
keywords mobile agentsGUI groundingvision-language modelsdecider-grounder architecturerecord-replay cachetask completion benchmarkDAG milestone evaluationGRPO curriculum training
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Mobile-agent research usually pits one large vision-language model against a benchmark; this paper instead builds and evaluates a full stack around a small model pair. Its central claim is that a 7B 'decider' that chooses the next action, a 3B 'grounder' that locates the target UI element, a record-replay cache, and a DAG-based benchmark together yield the best task-completion rates measured on real Chinese apps—higher than much larger general-purpose models and than a leading specialized GUI agent. The authors also report that the AgentRR replay mechanism reuses 60–85% of actions under realistic user workload distributions and cuts end-to-end latency by 2–3x. If true, the paper shows that specialization and caching can substitute for raw model scale in mobile device control.

Core claim

On the paper's own terms, the discovery is that a two-model architecture—a Planner/Decider/Grounder trio where the Decider outputs high-level action descriptions and the Grounder maps them to bounding boxes—can be trained by two-stage GRPO, with the trained Grounder acting as a reward model for the Decider, to outperform all tested alternatives on the MobiFlow benchmark. MobiFlow scores tasks not by whether the final state matched one script but by whether the agent hit milestone events arranged in a DAG, with multi-level checkers and manual review of failures. The same system adds ActTree, a tree of recorded UI-state transitions, so that on repeated tasks a lightweight latent-memory model r

What carries the argument

The load-bearing object is the Decider–Grounder pair, trained so that the Grounder's bounding-box prediction doubles as an outcome reward for the Decider; the second object is ActTree, a tree of UI states and actions that lets a tiny embedding/reranker decide whether to replay a cached prefix. Together they separate what to do (language-level decisions) from where to do it (pixel-level grounding), and separate fast routine execution from slow model inference.

Load-bearing premise

The ranking depends on MobiFlow's hand-curated tasks, milestone definitions, and manual failure review being a fair sample of real phone use; if those tasks were chosen around what the decider/grounder pipeline handles well, the reported advantage may not generalize to everyday app use.

What would settle it

Have annotators who did not design the system write a fresh set of tasks in the same Chinese apps sampled from real usage logs, run the same models, and compare completion and termination; if a frontier general-purpose model or the specialized GUI agent wins, the paper's main claim is overturned.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the MobiFlow numbers hold, a 7B/3B pair can beat models an order of magnitude larger on concrete device tasks, so mobile agents do not need to wait for next-generation frontier models.
  • Record-replay acceleration works because real user requests follow a power-law distribution; reuse of 60–85% of actions means deployed agents can approach API-agent latency while keeping GUI generality.
  • DAG-based milestones allow partial credit and multiple valid trajectories, making it feasible to benchmark third-party apps whose success criteria are not scriptable.
  • Using the Grounder as a reward model gives a dense training signal for click decisions without requiring manual action-labels at RL time.
  • Training-time self-evolution, where failed traces are corrected and folded back into data, implies the agent improves on its own mistakes after each test round.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The benchmark's neutrality is testable: an independent group could define new tasks sampled from real user queries in the same apps, and if the ranking flips, the advantage is partly benchmark-specific.
  • The paper does not directly compare with open-weight agents trained at larger scale; the natural next experiment is to apply the same decider-grounder plus ActTree recipe to larger base VLMs.
  • ActTree-style prefix reuse might transfer to web and desktop GUI agents, where task repetition and UI stability are similar; the 99% replay correctness claim provides a concrete target for such ports.
  • Because GPT and Gemini sometimes used AI search to bypass steps, future benchmarks should include apps without AI-assisted search to separate planning skill from search shortcutting.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper presents MobiAgent, a full-stack mobile agent system with three components: MobiMind agent models (Planner 4B, Decider 7B, Grounder 3B), the AgentRR record-replay acceleration framework (ActTree plus latent memory models), and the MobiFlow DAG-based benchmark. The decider and grounder are post-trained from Qwen2.5-VL using SFT followed by two-stage curriculum GRPO, with a grounder-as-reward-model scheme and a self-evolution loop. The paper claims state-of-the-art task completion versus GPT-5, Gemini-2.5 Pro, and UI-TARS on the MobiFlow benchmark, as well as 2-3x latency improvement and over 99% replay correctness under simulated user task distributions. The principal evidence is Figure 5 and Figure 6 plus a few textual counts; no per-model score table, confidence intervals, total task counts, or external validation are provided.

Significance. If the results hold, the paper would be a useful systems contribution: the planner/decider/grounder decomposition, the record-replay cache with invalidation and eviction, the DAG-based benchmark supporting multiple valid trajectories and partial credit, and the low-cost trajectory collection pipeline are all potentially valuable. Strengths include the explicit training equations, the detailed ActTree description, and the multi-level verification design of MobiFlow. However, the headline SOTA claim is only as strong as MobiFlow, and MobiFlow is author-curated, unreleased, and not cross-checked against any public benchmark. The paper itself cautions that absolute MobiFlow scores are 'primarily for reference' (Section 5.3). The evidence is therefore not yet sufficient to establish superiority over GPT-5, Gemini-2.5 Pro, or UI-TARS, despite the reported numbers being plausible.

major comments (4)
  1. [Section 5.3 / Section 6.1 / Figure 5] The central SOTA claim is measured solely on MobiFlow, an author-designed and unreleased benchmark. Section 5.3 states that the authors 'meticulously select tasks that can be correctly executed in the current environment' and that absolute scores are 'primarily for reference'; Section 6.1 reports manual verification of failures. These choices introduce selection bias that directly affects the comparison against GPT-5, Gemini-2.5 Pro, and UI-TARS. To make the claim load-bearing, the authors need to release the MobiFlow task set, DAG configurations, raw per-agent traces, per-task scores, and a precise task count, and ideally cross-evaluate on an established public benchmark. Without this, the reported superiority over the named baselines is not independently established.
  2. [Section 3.2 / Section 5.3] There is a potential train/evaluation overlap that is not addressed. The self-evolution procedure in Section 3.2 collects new tasks at test time, corrects failed traces, and merges them into the training data. Simultaneously, the MobiFlow task templates are authored by the same group that designs the decider/grounder and its output format. The paper does not state whether any MobiFlow tasks or their templates were seen during training, self-evolution, or data collection. If the same or paraphrased tasks appear in both training and evaluation, the SOTA comparison is circular. The revision should state the disjointness criterion explicitly and ideally pre-register the MobiFlow DAGs before model development.
  3. [Section 6.1 / Section 6.2] The quantitative evidence is under-specified. The paper says 'we generated three real-world test cases for each type of template' (Section 6.1) but does not report the total number of templates, applications, or tasks, nor per-model scores, confidence intervals, or error bars in Figure 5. The statement that 'GPT failed to properly terminate tasks in 11 application categories' lacks a definition of category, termination penalty, and denominator. Similarly, Section 6.2 reports action replay rates and 'correctness of action replay exceeds 99%' under uniform and power-law distributions, but the distribution construction, task corpus, and replay-error definitions are not given. Providing a full results table and the underlying task lists is necessary for reproducibility.
  4. [Section 4 / Section 6.2] The incremental contribution over prior work [6] needs clarification. AgentRR is presented as a core component of MobiAgent, but reference [6] appears to be the same authors' earlier 'Get Experience from Practice' paper. The reader cannot tell which parts of the ActTree, latent memory, and replay framework are new in this manuscript and which are inherited. This matters for the claim that the 2-3x speedup is a contribution of the present system. The revision should state the delta relative to [6] and report which experimental results are newly produced here.
minor comments (4)
  1. [Section 3.2 / Section 4.4] Several hyperparameters are left unspecified: the IoU threshold β and reward weight α in Eq. (1), the embedding similarity threshold τ1, the reranking threshold τ2, and the InfoNCE temperature τ in Section 4.4. Include the chosen values and, ideally, a sensitivity analysis.
  2. [Section 2.1 / Section 3.1] The Wait action is defined as Wait(sec: int) in Section 2.1, but the decider prompt in Section 3.1 says the wait action has no parameters and waits for 1 second. This discrepancy should be resolved.
  3. [Throughout] There are several typos and inconsistent terms: 'Secion' (Section 2.3), 'desciptions' (Section 2.3), 'reusablity' (Section 4.2), 'single step decison' (Section 2.3). The paper also alternates between 'Grounder' and 'Actor' (the latter appears in Section 4.5.1).
  4. [Figure 5 / Figure 6] The bar charts do not show per-run variability, sample counts, or statistical significance. Adding error bars or a companion table with raw scores and counts would improve interpretability.

Circularity Check

0 steps flagged

No formal circularity; the paper's claims are empirical and the benchmark, while author-designed, does not reduce to the method by construction.

full rationale

The central SOTA claim is an empirical comparison on MobiFlow, an author-built benchmark. The benchmark's DAG milestones are defined from task descriptions and independent checkers, not from MobiAgent's outputs, so the score is not definitionally equal to the method. Training rewards (IoU/center for grounder, grounder-as-RM for decider) are standard reward models fitted to human-annotated bboxes/actions, and the AgentRR replay rates are measured from recorded traces; no equation in the paper reduces a prediction to a fitted input. The paper does reuse the authors' prior AgentRR work [6] as one of three components, but it presents its own experiments in Sec. 6.2 and does not use the citation to derive the task-completion result. The main weakness is external validity: Sec. 5.3 says 'we meticulously select tasks that can be correctly executed in the current environment' and Sec. 6.1 says 'we manually verified all failed tasks,' so the benchmark is curated and adjudicated by the same authors and is not cross-checked against a public benchmark. That is a benchmarking limitation, not a circular derivation. Hence no circularity steps are identified.

Axiom & Free-Parameter Ledger

6 free parameters · 7 axioms · 3 invented entities

The central claim rests on the authors' post-training recipe, their internal data collection with Gemini-2.5, and an unreleased benchmark. No equation is derived from first principles; the paper is an empirical systems contribution whose conclusions depend on the validity of self-created training and evaluation artifacts.

free parameters (6)
  • IoU reward threshold beta = not stated
    In the grounding GRPO reward, Riou = alpha if IoU > beta else 0; beta determines the click precision reward and is chosen by hand, not derived.
  • IoU reward weight alpha = not stated
    Blends the IoU and center-point rewards in the grounding GRPO reward; no value or tuning procedure is reported.
  • Embedding similarity threshold tau1 = not stated
    Controls stage-1 action-reuse decisions in AgentRR; it is chosen manually and directly affects the reported replay rate and correctness.
  • Reranking confidence threshold tau2 = not stated
    Controls stage-2 reuse decisions; the reported >99% replay correctness and 30-85% replay rates depend on this threshold.
  • InfoNCE temperature tau = not stated
    Temperature in the contrastive loss for the task embedding model; standard but unstated.
  • Curriculum difficulty labels = binary labels set by predefined rules
    Difficult/easy assignment per action is chosen with predefined rules and manual importance; the rules are not specified and this shapes decider training.
axioms (7)
  • domain assumption Qwen2.5-VL-7B/3B provide adequate visual-linguistic priors for post-training
    The MobiMind decider and grounder are post-trained from these checkpoints; if the base VLM cannot perceive GUI screenshots, the whole pipeline fails.
  • domain assumption Gemini-2.5-generated reasoning and action primitives are reliable ground truth
    Section 2.2 uses Gemini-2.5 to reconstruct reasoning for every recorded action; labeling errors propagate into SFT and GRPO training.
  • domain assumption The action recording tool and OmniParser bbox regeneration capture the true interactive UI
    Section 2.1 relies on XML-based bboxes with OmniParser fallback; missing or stale bboxes would mislabel coordinates.
  • domain assumption MobiFlow milestone checkers (text, OCR, icon, LLM judge) correctly verify task completion
    Section 5.2 defines the verification mechanism, but no accuracy or inter-rater reliability of the checkers is reported; the entire score depends on it.
  • domain assumption Latent memory models trained with prefix-derived positive/negative samples generalize to unseen tasks
    Section 4.4.3 defines positives as tasks sharing an l-step prefix; this assumes prefix overlap is the right signal for safe action reuse.
  • domain assumption Separate GRPO training of grounder then decider approximates joint multi-agent optimization
    Section 3.2 simplifies multi-agent RL into two independent stages; no analysis shows this preserves joint-policy optimality.
  • domain assumption Manual verification of failed tasks and repeated testing remove environment confounds
    Section 6.1 states manual verification and repeated tests; no protocol, inter-rater agreement, or predefined handling rules are given, so the claim relies on the authors' judgment.
invented entities (3)
  • ActTree no independent evidence
    purpose: Tree-structured cache of UI states and reusable action transitions for record-replay acceleration
    An engineered data structure; its benefits are only shown in the authors' own simulation and app tests.
  • Latent Memory model (task embedding + reranker) no independent evidence
    purpose: Decides how many prefix actions can be reused by comparing task similarity
    No external checkpoint, dataset, or public evaluation is provided; it is trained and evaluated only within this paper.
  • MobiFlow benchmark and task suite no independent evidence
    purpose: New DAG-based evaluation of mobile agents on Chinese apps
    The benchmark is not released and no cross-validation against existing benchmarks is provided.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of MobiAgent: A Systematic Framework for Customizable Mobile Agents." pith.science (2026). https://pith.science/paper/VY36LQCY

@misc{pith2026250900531,
  author       = {Pith},
  title        = {Pith review of: MobiAgent: A Systematic Framework for Customizable Mobile Agents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VY36LQCY}},
  note         = {Machine review of arXiv:2509.00531}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

With the rapid advancement of Vision-Language Models (VLMs), GUI-based mobile agents have emerged as a key development direction for intelligent mobile systems. However, existing agent models continue to face significant challenges in real-world task execution, particularly in terms of accuracy and efficiency. To address these limitations, we propose MobiAgent, a comprehensive mobile agent system comprising three core components: the MobiMind-series agent models, the AgentRR acceleration framework, and the MobiFlow benchmarking suite. Furthermore, recognizing that the capabilities of current mobile agents are still limited by the availability of high-quality data, we have developed an AI-assisted agile data collection pipeline that significantly reduces the cost of manual annotation. Compared to both general-purpose LLMs and specialized GUI agent models, MobiAgent achieves state-of-the-art performance in real-world mobile scenarios.

Figures

Figures reproduced from arXiv: 2509.00531 by Cheng Zhang, Dong Du, Erhu Feng, Haibo Chen, Jiahui Sun, Wangbo Gong, Xi Zhao, Yisheng Zhao, Yubin Xia, Zhichao Hua.

Figure 1
Figure 1. Figure 1: Data Collection Pipeline with Agent Self-evolving crucial for training intelligent and robust agent models. To address this issue, we propose a VLM￾based reasoning reconstruction approach which leverages Gemini-2.5 to reconstruct the agent’s reasoning process based on original trajectories. Specifically, we prompt the VLM to examine each action in the trajectory and provide a detailed reasoning, simulating… view at source ↗
Figure 2
Figure 2. Figure 2: Multi-Agent Architecture using the AgentRR Framework latent memory, enabling swift execution of familiar actions while preserving the flexibility required for new situations. 4.2 Prefix Reusablity of GUI tasks We observe the prefix reusablity of GUI tasks, which relies on the following insights: • I1: The execution trajectories of similar tasks share identical prefixes. • I2: Since the agent’s planning can… view at source ↗
Figure 3
Figure 3. Figure 3: Construction of the ActTree Structure During Mobile Task Execution and (2) a task list Te = {t1, ..., tk}, which records all historical tasks that executed action Ae to transition from UIu to UIv. The construction of the ActTree proceeds incrementally during the task executions. When a new action is performed, we first create a new node v to represent the resulting UI state: UIv, along with a corresponding… view at source ↗
Figure 4
Figure 4. Figure 4: The overall architecture of MobiFlow in the DAG are awarded the full score, while tasks that accomplish only a subset of milestone events may receive partial credit, as specified by user-defined criteria. In real-world scenarios, application responses are highly dependent on factors such as application version, environment, and interaction patterns. Moreover, there is often a lack of systematic interfaces … view at source ↗
Figure 5
Figure 5. Figure 5: Average Task Completion Scores of Different Agent Models Under Realistic Workloads (MobiFlow): The performance is shown for (a) all tasks combined; (b) easy tasks; (c) and hard tasks. as travel, shopping, entertainment and social networking. For each application, we defined a set of test case templates with multiple difficulty levels, where each template consists of milestone events structured in the form … view at source ↗
Figure 6
Figure 6. Figure 6: Action Replay Rate with AgentRR under Different Distributions. shopping, AgentRR achieves an average performance improvement of 2 to 3 times compared to baseline approaches. 7 Conclusion In this paper, we address the challenges of building, evaluating and deploying mobile agent by introducing a comprehensive, full-stack solution encompassing the MobiMind model series, the AgentRR acceleration framework, an… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 9 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. FlexServe: A Fast and Secure LLM Serving System for Mobile Devices with Flexible Resource Isolation

    cs.CR 2026-06 unverdicted novelty 7.0

    FlexServe decouples access and management of secure resources in TrustZone to enable efficient LLM inference on mobiles, reporting 10.05X TTFT speedup over basic strawman designs and 2.44X over optimized ones.

  2. FlexServe: A Fast and Secure LLM Serving System for Mobile Devices with Flexible Resource Isolation

    cs.CR 2026-06 unverdicted novelty 7.0

    FlexServe introduces recallable secure memory and NPU to enable cooperative secure LLM inference on mobile devices, reporting 10.05X TTFT speedup over a basic TrustZone strawman.

  3. FlexServe: A Fast and Secure LLM Serving System for Mobile Devices with Flexible Resource Isolation

    cs.CR 2026-03 conditional novelty 6.5

    Page-granular Flex-Mem and switchable Flex-NPU cut TrustZone LLM TTFT by ~10× vs a CMA strawman and ~2.4× vs a pipelined secure-NPU strawman on RK3588.

  4. DeltaBox: Scaling Stateful AI Agents with Millisecond-Level Sandbox Checkpoint/Rollback

    cs.OS 2026-05 unverdicted novelty 6.0

    DeltaBox achieves millisecond-level checkpoint (14ms) and rollback (5ms) for AI agent sandboxes by layering file states and using incremental process dumps to exploit similarity between consecutive checkpoints.

  5. DeltaBox: Scaling Stateful AI Agents with Millisecond-Level Sandbox Checkpoint/Rollback

    cs.OS 2026-05 unverdicted novelty 6.0

    DeltaBox achieves 14 ms checkpoint and 5 ms rollback for AI agent sandboxes via layered DeltaFS and incremental DeltaCR mechanisms that exploit similarity between consecutive states.

  6. EdgeFlow: Fast Cold Starts for LLMs on Mobile Devices

    cs.OS 2026-04 unverdicted novelty 6.0

    EdgeFlow reduces mobile LLM cold-start latency up to 4.07x versus llama.cpp, MNN, and llm.npu by NPU-aware adaptive quantization, SIMD-friendly packing, and synergistic granular CPU-NPU pipelining at comparable accuracy.

  7. FlexServe: A Fast and Secure LLM Serving System for Mobile Devices with Flexible Resource Isolation

    cs.CR 2026-03 unverdicted novelty 6.0

    FlexServe achieves up to 10x faster time-to-first-token for secure LLM inference on mobile devices by using flexible resource isolation in TrustZone compared to standard approaches.

  8. Mind the Gap: Action Rebinding Attacks against Android GUI Agents

    cs.CR 2026-01 conditional novelty 6.0

    A zero-permission Android app can redirect a GUI agent's planned tap to a different app by switching the foreground during the agent's reasoning delay.

  9. SE-GA: Memory-Augmented Self-Evolution for GUI Agents

    cs.LG 2026-05 unverdicted novelty 5.0

    SE-GA combines Test-Time Memory Extension for dynamic context retrieval with Memory-Augmented Self-Evolution training to reach 89.0% on ScreenSpot and 75.8% on AndroidControl-High.

Reference graph

Works this paper leans on

35 extracted references · 10 canonical work pages · cited by 6 Pith papers

  1. [1]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Hang Zhang, Zhibo Yang, Haiyang Xu, and Junyang Lin. 2025. Qwen2.5-VL Technical Rep...

  2. [2]

    Yoshua Bengio, Jérôme Louradour, Ronan Collobert, and Jason Weston. 2009. Curriculum learning. In Proceedings of the 26th Annual International Conference on Machine Learning (Montreal, Quebec, Canada) (ICML ’09). Association for Computing Machinery, New York, NY , USA, 41–48. https://doi.org/10.1145/1553374.1553380

  3. [3]

    G. Bradski. 2000. The OpenCV Library. Dr. Dobb’s Journal of Software Tools(2000)

  4. [4]

    Yuxiang Chai, Hanhao Li, Jiayu Zhang, Liang Liu, Guangyi Liu, Guozhi Wang, Shuai Ren, Siyuan Huang, and Hongsheng Li. 2025. A3: Android Agent Arena for Mobile GUI Agents. arXiv:2501.01149 [cs.AI] https://arxiv.org/abs/2501.01149

  5. [5]

    Kanzhi Cheng, Qiushi Sun, Yougang Chu, Fangzhi Xu, Yantao Li, Jianbing Zhang, and Zhiyong Wu. 2024. SeeClick: Harnessing GUI Grounding for Advanced Visual GUI Agents. arXiv:2401.10935 [cs.HC] https://arxiv.org/abs/2401.10935

  6. [6]

    Erhu Feng, Wenbo Zhou, Zibin Liu, Le Chen, Yunpeng Dong, Cheng Zhang, Yisheng Zhao, Dong Du, Zhichao Hua, Yubin Xia, and Haibo Chen. 2025. Get Experience from Practice: LLM Agents with Record & Replay. arXiv:2505.17716 [cs.LG] https://arxiv.org/abs/ 2505.17716 14

  7. [7]

    Google. 2025. Gemini 2.5 Pro Best for coding and highly complex tasks. https://deepmind. google/models/gemini/pro/

  8. [8]

    Boyu Gou, Ruohan Wang, Boyuan Zheng, Yanan Xie, Cheng Chang, Yiheng Shu, Huan Sun, and Yu Su. 2025. Navigating the Digital World as Humans Do: Universal Visual Grounding for GUI Agents. In The Thirteenth International Conference on Learning Representations . https://openreview.net/forum?id=kxnoqaisCT

  9. [9]

    Wenyi Hong, Weihan Wang, Qingsong Lv, Jiazheng Xu, Wenmeng Yu, Junhui Ji, Yan Wang, Zihan Wang, Yuxuan Zhang, Juanzi Li, Bin Xu, Yuxiao Dong, Ming Ding, and Jie Tang

  10. [10]

    Raghav Kapoor, Yash Parag Butala, Melisa Russak, Jing Yu Koh, Kiran Kamble, Waseem Alshikh, and Ruslan Salakhutdinov. 2024. OmniACT: A Dataset and Benchmark for Enabling Multimodal Generalist Autonomous Agents for Desktop and Web. arXiv:2402.17553 [cs.AI] https://arxiv.org/abs/2402.17553

  11. [11]

    Juyong Lee, Taywon Min, Minyong An, Dongyoon Hahm, Haeone Lee, Changyeon Kim, and Kimin Lee. 2024. Benchmarking Mobile Device Control Agents Across Diverse Configurations. arXiv preprint arXiv:2404.16660 (2024)

  12. [12]

    Sunjae Lee, Junyoung Choi, Jungjae Lee, Munim Hasan Wasi, Hojun Choi, Steve Ko, Sangeun Oh, and Insik Shin. 2024. MobileGPT: Augmenting LLM with Human-like App Memory for Mobile Task Automation. In Proceedings of the 30th Annual International Conference on Mobile Computing and Networking (Washington D.C., DC, USA) (ACM MobiCom ’24). Association for Comput...

  13. [13]

    Kaixin Li, Ziyang Meng, Hongzhan Lin, Ziyang Luo, Yuchen Tian, Jing Ma, Zhiyong Huang, and Tat-Seng Chua. 2025. ScreenSpot-Pro: GUI Grounding for Professional High-Resolution Computer Use. arXiv:2504.07981 [cs.CV] https://arxiv.org/abs/2504.07981

  14. [14]

    Wei Li, William Bishop, Alice Li, Chris Rawles, Folawiyo Campbell-Ajala, Divya Tya- magundlu, and Oriana Riva. 2024. On the Effects of Data Scale on UI Control Agents. arXiv:2406.03679 [cs.AI] https://arxiv.org/abs/2406.03679

  15. [15]

    Kevin Qinghong Lin, Linjie Li, Difei Gao, Zhengyuan Yang, Shiwei Wu, Zechen Bai, Weixian Lei, Lijuan Wang, and Mike Zheng Shou. 2024. ShowUI: One Vision-Language-Action Model for GUI Visual Agent. arXiv:2411.17465 [cs.CV] https://arxiv.org/abs/2411.17465

  16. [16]

    Quanfeng Lu, Wenqi Shao, Zitao Liu, Lingxiao Du, Fanqing Meng, Boxuan Li, Botong Chen, Siyuan Huang, Kaipeng Zhang, and Ping Luo. 2025. GUIOdyssey: A Comprehensive Dataset for Cross-App GUI Navigation on Mobile Devices. arXiv:2406.08451 [cs.CV] https: //arxiv.org/abs/2406.08451

  17. [17]

    Yadong Lu, Jianwei Yang, Yelong Shen, and Ahmed Awadallah. 2024. OmniParser for Pure Vision Based GUI Agent. arXiv:2408.00203 [cs.CV] https://arxiv.org/abs/2408. 00203

  18. [18]

    OpenAI. 2025. GPT-5 is here. https://openai.com/gpt-5/

  19. [19]

    Jiayi Pan, Yichi Zhang, Nicholas Tomlin, Yifei Zhou, Sergey Levine, and Alane Suhr. 2024. Autonomous Evaluation and Refinement of Digital Agents. arXiv:2404.06474 [cs.AI]

  20. [20]

    Yujia Qin, Yining Ye, Junjie Fang, Haoming Wang, Shihao Liang, Shizuo Tian, Junda Zhang, Jiahao Li, Yunxin Li, Shijue Huang, Wanjun Zhong, Kuanye Li, Jiale Yang, Yu Miao, Woyu Lin, Longxiang Liu, Xu Jiang, Qianli Ma, Jingyu Li, Xiaojun Xiao, Kai Cai, Chuang Li, Yaowei Zheng, Chaolin Jin, Chen Li, Xiao Zhou, Minchao Wang, Haoli Chen, Zhaojian Li, Haihua Ya...

  21. [21]

    Christopher Rawles, Sarah Clinckemaillie, Yifan Chang, Jonathan Waltz, Gabrielle Lau, Mary- beth Fair, Alice Li, William Bishop, Wei Li, Folawiyo Campbell-Ajala, Daniel Toyama, Robert Berry, Divya Tyamagundlu, Timothy Lillicrap, and Oriana Riva. 2025. AndroidWorld: A Dynamic Benchmarking Environment for Autonomous Agents. arXiv:2405.14573 [cs.AI] https://...

  22. [22]

    Christopher Rawles, Alice Li, Daniel Rodriguez, Oriana Riva, and Timothy Lillicrap. 2023. Android in the wild: a large-scale dataset for android device control. In Proceedings of the 37th International Conference on Neural Information Processing Systems (New Orleans, LA, USA) (NIPS ’23). Curran Associates Inc., Red Hook, NY , USA, Article 2609, 21 pages

  23. [23]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y . K. Li, Y . Wu, and Daya Guo. 2024. DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models. arXiv:2402.03300 [cs.CL] https://arxiv.org/abs/2402.03300

  24. [24]

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. 2019. Representation Learning with Contrastive Predictive Coding. arXiv:1807.03748 [cs.LG] https://arxiv.org/abs/1807. 03748

  25. [25]

    Junyang Wang, Haiyang Xu, Haitao Jia, Xi Zhang, Ming Yan, Weizhou Shen, Ji Zhang, Fei Huang, and Jitao Sang. 2024. Mobile-Agent-v2: Mobile Device Operation Assistant with Effective Navigation via Multi-Agent Collaboration. arXiv:2406.01014 [cs.CL] https: //arxiv.org/abs/2406.01014

  26. [26]

    Hao Wen, Yuanchun Li, Guohong Liu, Shanhui Zhao, Tao Yu, Toby Jia-Jun Li, Shiqi Jiang, Yun- hao Liu, Yaqin Zhang, and Yunxin Liu. 2024. AutoDroid: LLM-powered Task Automation in Android. In Proceedings of the 30th Annual International Conference on Mobile Computing and Networking (Washington D.C., DC, USA)(ACM MobiCom ’24). Association for Computing Ma- c...

  27. [27]

    Zhiyong Wu, Zhenyu Wu, Fangzhi Xu, Yian Wang, Qiushi Sun, Chengyou Jia, Kanzhi Cheng, Zichen Ding, Liheng Chen, Paul Pu Liang, and Yu Qiao. 2024. OS-ATLAS: A Foundation Action Model for Generalist GUI Agents. arXiv:2410.23218 [cs.CL] https://arxiv.org/ abs/2410.23218

  28. [28]

    Mingzhe Xing, Rongkai Zhang, Hui Xue, Qi Chen, Fan Yang, and Zhen Xiao. 2024. Understand- ing the Weakness of Large Language Model Agents within a Complex Android Environment. arXiv:2402.06596 [cs.AI] https://arxiv.org/abs/2402.06596

  29. [29]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. ReAct: Synergizing Reasoning and Acting in Language Models. In International Conference on Learning Representations (ICLR)

  30. [30]

    Chi Zhang, Zhao Yang, Jiaxuan Liu, Yucheng Han, Xin Chen, Zebiao Huang, Bin Fu, and Gang Yu. 2023. AppAgent: Multimodal Agents as Smartphone Users. arXiv:2312.13771 [cs.CV] https://arxiv.org/abs/2312.13771

  31. [31]

    Danyang Zhang, Zhennan Shen, Rui Xie, Situo Zhang, Tianbao Xie, Zihan Zhao, Siyuan Chen, Lu Chen, Hongshen Xu, Ruisheng Cao, and Kai Yu. 2024. Mobile-Env: Building Qualified Evaluation Benchmarks for LLM-GUI Interaction. arXiv:2305.08144 [cs.AI] https: //arxiv.org/abs/2305.08144

  32. [32]

    Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. 2025. Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models. arXiv:2506.05176 [cs.CL] https://arxiv.org/abs/2506.05176

  33. [33]

    Zhuosheng Zhang and Aston Zhang. 2024. You Only Look at Screens: Multimodal Chain-of- Action Agents. arXiv:2309.11436 [cs.CL] https://arxiv.org/abs/2309.11436

  34. [34]

    Boyuan Zheng, Boyu Gou, Jihyung Kil, Huan Sun, and Yu Su. 2024. GPT-4V(ision) is a Generalist Web Agent, if Grounded. In Forty-first International Conference on Machine Learning. https://openreview.net/forum?id=piecKJ2DlB 16

  35. [2024]

    arXiv:2312.08914 [cs.CV] https://arxiv.org/abs/2312.08914

    CogAgent: A Visual Language Model for GUI Agents. arXiv:2312.08914 [cs.CV] https://arxiv.org/abs/2312.08914

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.