Zeyu Zhang

Chinese Academy of Sciences · China

Unclaimed · AI-structured profileOpenAlex

Early-career CAS researcher linking real-device GUI agents, learning-based manipulation, and LLM deep-research agents

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-17) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

Output measures publication volume and productivity over time. This axis is not yet scored: the underlying production data in our snapshot is insufficient, so no output estimate is shown here rather than a guessed value.

Impact—

Impact captures citation-based influence, such as total citations, h-index trajectory, and field-weighted attention. The required citation data is not yet available for this profile, so no impact score is computed or estimated here.

Focus0/100

Focus measures concentration of research topics. The recorded score of 0 reflects maximal breadth: the field distribution splits evenly at 0.333 across robot manipulation, multimodal GUI agents, and LLM-based agents, while OpenAlex topics scatter across multimodal machine learning (6), reinforcement learning in robotics (5), robot manipulation learning (4), 3D shape modeling (4), and neural network applications (3) — deliberate multi-front work rather than a single-thread agenda.

Collaboration82/100

With 59 distinct co-authors recorded by OpenAlex and a collaboration score of 82, Zeyu Zhang works in strongly team-based settings. Twelve collaborators recur most often — including Sixu Yan, Shikang Wang, and Hangxin Liu — spanning robotics and vision, consistent with the multi-team scale typical of technical reports like BlueLM-GUI. All three cataloged papers list the researcher in middle author positions without corresponding-author status.

Activity83/100

Activity measures recent publication rhythm. The score of 83 is grounded in 16 works over the last three years, with OpenAlex counts by year placing all 16 in 2026 — a concentrated burst spanning GUI agents, grasping, and deep-research agents. For a record with an h-index of 1, this tempo indicates a researcher in a highly productive early phase.

Reproducibility32/100

Reproducibility summarizes Evolve's readiness assessments across 3 reviewed papers, averaging 32/100. Two papers rate low: the grasping work ships no code, statistical tests, or real-world baselines, and BlueLM-GUI's evaluation is partly self-referential with a confounded ablation. APEx rates medium but is criticized for cherry-picking a weak baseline behind its 14.7-point GPT-5.4 gap.

Research summary

Zeyu Zhang is an early-stage researcher at the Chinese Academy of Sciences whose 2026 output concentrates on three complementary branches of embodied and agentic AI. OpenAlex records 16 works in the last three years — all counted in 2026 — with 59 distinct co-authors and an h-index of 1 (i10-index 0), a profile consistent with a researcher just entering the literature. The three papers Evolve has reviewed span robot manipulation, multimodal GUI agents, and LLM-based agents in equal measure (0.333 each in the field distribution). The BlueLM-GUI technical report builds a real-device, self-improving flywheel for mobile GUI agents, reporting 87.4 on its own benchmark and 84.9 on AndroidWorld; the grasping work composes foundation priors with generalizable, hand-agnostic grasp synthesis for vision-language grasping; APEx distills agent procedural experience into skills with test-time reinforcement learning for deep-research question answering. Across the trio, reviewers note strong empirical results alongside reproducibility concerns — two papers rated low and one medium — including missing code, self-referential evaluation, and selectively chosen baselines. Topic data (multimodal machine learning, reinforcement learning in robotics, 3D shape modeling) reinforce a learning-centric, systems-building style. This is an early-stage profile: citation-based impact measures are not yet available.

Collaboration pattern

Zeyu Zhang's record is strongly collaborative: OpenAlex logs 59 distinct co-authors, and twelve recur most often, including Sixu Yan, Shikang Wang, Binhua Huang, and Hangxin Liu. The collaborator pool spans robotics, vision, and language-model work, mirroring the three-way field split. On all three cataloged papers the researcher appears in a middle author position and is never the corresponding author, suggesting a hands-on contributor within larger teams rather than a project lead so far.

Research keywords

mobile GUI agentsvision-language-action modelsagentic reinforcement learningGRPOrobot manipulationgrasp synthesisdexterous manipulationcross-embodiment generalizationdeep research agentsprocedural skill distillationtest-time reinforcement learning

Highlights

Real-device flywheel for self-improving mobile GUI agents
The BlueLM-GUI technical report trains vision-language-action agents on real phones, reporting 87.4 on its own benchmark and 84.9 on AndroidWorld; Evolve credits the flywheel design but flags self-referential evaluation and a confounded ablation.
Composable foundation priors for vision-language grasping
This work decouples foundation-model priors from a hand-agnostic, generalizable grasp synthesis module, delivering strong sim-to-real manipulation results — though Evolve notes the absence of code, statistical tests, and real-world baselines.
Distilling agent experience into skills for deep-research QA
APEx converts procedural agent experience into reusable skills combined with test-time reinforcement learning for adaptive deep-research question answering; the SOTA claim is credible, but the 14.7-point GPT-5.4 gap leans on a weak baseline.

Timeline

2026Published the BlueLM-GUI technical report on a real-device-centric, self-improving flywheel for mobile GUI agents.
2026Published work on adaptive vision-language grasping combining composable foundation priors with generalizable grasp synthesis.
2026Published APEx on distilling agent procedural experience for adaptive deep-research question answering.
2026OpenAlex recorded 16 works in the last three years — all in 2026 — with 59 distinct co-authors.

Papers in the Evolve catalog

Frequent co-authors

Sixu Yan×1 Shikang Wang×1 Binhua Huang×1 Xuanlai Tang×1 Guohua Fan×1 Fan Huang×1 Haoxuan Li×1 Yongkang Li×1 Yuhan Li×1 Bencheng Liao×1 Wenyu Liu×1 Hangxin Liu×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

Bibliometric indicators are computed from OpenAlex only (identity: openalex); no ORCID is linked, so the profile is name-resolved from public bibliographic data.
"Zeyu Zhang" is a common name; disambiguation relies on OpenAlex author clustering and may misattribute or omit papers.
Output and impact axes are unscored because citation data is insufficient; the raw record shows only h-index 1 and i10-index 0, consistent with a very early career.
All three reviewed papers are 2026 arXiv preprints with middle authorship and no corresponding role; peer-review status and independent citation impact are not yet established.