Robotics researcher at Midea Group working on world action models and vision-language-action manipulation
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) 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 metrics are not yet available for scoring: the underlying record shows an h-index of 0 and an i10-index of 0, so this axis is left unscored rather than estimated. Once complete, it would measure cumulative publication volume and per-year productivity of this researcher.
Impact—
Impact cannot yet be scored: the record offers no citation aggregates beyond a recorded h-index of 0 and i10-index of 0, and the reviewed papers are recent preprints. This axis would measure citation counts and uptake of the work by other researchers.
Focus0/100
The focus score is 0, reflecting an evenly split profile: the field distribution assigns 0.5 to robot manipulation / embodied AI and 0.5 to robot learning / vision-language-action models. Topic counts (robot manipulation and learning, 4; reinforcement learning in robotics, 3; AI-based problem solving and planning, 2; multimodal machine learning applications, 2) show the two halves are closely adjacent, so the spread is shallow rather than scattered.
Collaboration55/100
Collaboration scores 55, based on 15 distinct co-authors in the recent three-year window. The co-author set is stable and team-based: Zitai Huang, Taiyi Su, Jianjun Zhang, Chong Ma, Tianbin Liu, Yi Xu, Weiyi Lu, and Hanli Wang recur across records, and both cataloged papers list Jian Zhu in a middle author position, consistent with a contributing member of a corporate robotics group rather than a solo lead.
Activity64/100
Activity scores 64. The indexed record contains 8 works in the recent three-year window, and counts by year place all 8 in 2026 — a concentrated burst rather than a sustained multi-year curve. Together with 15 distinct co-authors over the same period, this points to a currently active, team-driven pace centered on arXiv preprints in robot learning.
Reproducibility20/100
Reproducibility scores 20, computed from 2 assessed papers with a mean readiness of 20. Both cataloged papers are rated low: JEPA-WAM reports no task-agnostic control, no significance tests, and no released code, while MaskHarness-WAM lacks component ablations. Both claim strong real-robot gains, but external replication would currently be difficult without released artifacts.
Research summary
Jian Zhu is a robotics researcher at Midea Group (China) whose recent work centers on world action models and vision-language-action systems for real-robot manipulation. The Evolve catalog includes two 2026 arXiv preprints: JEPA-WAM, which augments world action models with text-to-image-generated visual instructions encoded by V-JEPA and reports strong real-robot wins over Fast-WAM, and MaskHarness-WAM, which uses a VLM-gated harness to re-ground target masks per subtask for long-horizon, identical-object sequential manipulation. In both papers Zhu holds a middle author position, consistent with a contributing member of a corporate robotics team rather than a solo project lead. Bibliometric records (OpenAlex identity) show 8 works in the recent three-year window, all dated 2026, with 15 distinct co-authors and a stable collaborator set including Zitai Huang, Taiyi Su, Jianjun Zhang, Chong Ma, Tianbin Liu, Yi Xu, Weiyi Lu, and Hanli Wang. The field distribution splits evenly (0.5/0.5) between robot manipulation/embodied AI and robot learning/VLA models, with reinforcement learning and multimodal learning as secondary themes. Citation-based metrics (h-index 0, i10-index 0) and reproducibility readiness (mean 20 across 2 assessed papers, both rated low) mean output and impact cannot yet be meaningfully scored; Evolve's verdicts note strong empirical results but missing controls, ablations, significance tests, and released code.
Collaboration pattern
Zhu's recent work is highly team-based: 15 distinct co-authors appear in the recent three-year OpenAlex window, with a recurring core of Zitai Huang, Taiyi Su, Jianjun Zhang, Chong Ma, Tianbin Liu, Yi Xu, Weiyi Lu, and Hanli Wang. On both cataloged 2026 papers he holds a middle author position without corresponding-author status, suggesting he contributes specialist expertise inside a Midea robotics group rather than leading papers as sole or senior author.
Research keywords
Highlights
JEPA-WAM: generated visual instructions for world action models
MaskHarness-WAM: instance-grounded harnessing for long-horizon manipulation
A concentrated 2026 push on WAM-based real-robot manipulation
Both reviewed papers come from the same 2026 research line on world action models, with Jian Zhu as a middle author on each. They sit within a broader burst of 8 OpenAlex-indexed works in 2026 alongside 15 distinct co-authors, indicating a fast-moving but early and not yet independently verified body of work.
Timeline
- 2026JEPA-WAM preprint released on arXiv, connecting text-to-image-generated visual instructions to world action models through V-JEPA latent representations.
- 2026MaskHarness-WAM preprint released on arXiv, targeting long-horizon robot manipulation via per-subtask re-grounding of instance masks.
- 2026OpenAlex-indexed output peaks at 8 works in this single year, with 15 distinct co-authors recorded across the recent three-year window.
Papers in the Evolve catalog
Frequent co-authors
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.
- Bibliometrics are drawn from OpenAlex only; the h-index and i10-index are recorded as 0, so the output and impact dimensions are unscored rather than assessed as weak.
- The profile is name-resolved from public bibliographic data (identity: openalex) with no ORCID link; Jian Zhu is a common name, so record-level attribution may be imperfect.
- Only two Evolve-reviewed papers are available, both 2026 arXiv preprints without peer review, so conclusions rest on a very small and early sample.
- The reproducibility score (20) is computed from just 2 assessed papers, both rated low readiness, making the axis highly sensitive to future additions.