Embodied AI researcher bridging decision-aligned world models and VLA robustness calibration for robot manipulation
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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—
No publication-volume data is available in this snapshot: output measures cumulative and per-year paper counts across venues. With only two reviewed 2026 arXiv papers recorded, this axis cannot be scored yet.
Impact—
Impact cannot be assessed: this axis tracks citations, citation velocity, and uptake of the two 2026 papers. No citation counts exist in the input data, so no score is estimated.
Focus0/100
Radar scores focus at 0 from an even 0.5/0.5 split across two field labels: robotics/embodied machine learning and robot learning/world models. In practice the two papers are topically adjacent — one builds a decision-aligned world model for planning, the other calibrates frozen VLA manipulation policies — so the low score reflects taxonomy splitting more than genuine scatter of research interests.
Collaboration55/100
Scored 55 from 15 distinct coauthors across the record. The frequent-coauthor list repeats Shuaijun Liu and Feiyang You, pointing to a small recurring core team, while names such as Chenglong Zhang, Shuyang Hao, and Ningxin Su broaden the circle. Both cataloged papers list Wu as a middle author and not corresponding, consistent with contributing inside larger teams rather than leading them.
Activity—
Activity (recent publication cadence and year-over-year consistency) lacks the underlying per-year record: only the two 2026 papers are dated, so no trend can be computed or scored from the available data.
Reproducibility55/100
Reproducibility averages 55 across the two papers scored (reproPapers 2), both rated medium readiness. Evolve's review of D-JEPA notes consistent simulation and real-robot gains but flags cohort selection, a tiny driving set, and no code release; BAS-VLA shows solid simulations under appearance shift with a reported 0% stale-success rate under semantic change, though preservation evidence is narrow.
Research summary
Chengyu Wu works on embodied AI and machine learning for robotics; this profile is at an early stage, name-resolved from public bibliographic data and keyed to ORCID 0000-0002-7467-2737, with no institution or country on record. Evolve's catalog holds two 2026 arXiv papers, both with Wu as a middle author and neither as corresponding author. The two works approach reliable robot decision-making from complementary sides. D-JEPA proposes a decision-aligned latent world model that re-ranks candidate futures through a bounded set transformer; Evolve notes consistent simulation and real-robot gains, alongside caveats about cohort selection, a tiny driving set, and no code release. BAS-VLA adds a task-semantic action-calibration layer on top of a frozen vision-language-action model: stable under appearance shift, but Evolve's semantic-breaking interventions drive stale success to 0%, with solid simulations yet narrow preservation evidence. Bibliometrically the footprint is narrow but coherent — the field distribution splits evenly (0.5/0.5) between robotics/embodied machine learning and robot learning/world models; the collaboration radar counts 15 distinct coauthors (score 55), with Shuaijun Liu and Feiyang You recurring; and reproducibility averages 55 across the two scored papers, both rated medium readiness. Citation, output-volume, and per-year activity data are absent, so impact, productivity, and activity cannot be scored; read this as a focused snapshot of a 2026-era research program rather than a mature bibliometric record.
Collaboration pattern
The record shows 15 distinct coauthors (collaboration score 55). Shuaijun Liu and Feiyang You appear twice in the frequent-coauthor list, marking the closest recurring partners; Shuyang Hao, Chenglong Zhang, Ningxin Su, Jingyao Cai, Xingwei Chen, Qifu Wen, and Xi Lin round out a broader rotation. In both cataloged papers Wu sits in the middle of the author list without corresponding-author status, suggesting a contributing role within larger robotics/ML teams.
Research keywords
Highlights
Decision-aligned re-ranking for latent world models
Frozen-VLA calibration that survives appearance but not semantic shift
A coherent 2026 program around reliable robot decision-making
Both 2026 papers attack the reliability of learned robot policies from complementary sides — better candidate evaluation inside world-model planning, and principled counterfactual evaluation of deployed VLA policies — splitting evenly (0.5/0.5) across the profile's two field labels.
Timeline
- 2026D-JEPA posted to arXiv: a decision-aligned latent world model that re-ranks candidate futures with a bounded set transformer (middle authorship, not corresponding).
- 2026BAS-VLA posted to arXiv: task-semantic action calibration for frozen vision-language-action models, stable under appearance shift with 0% stale success under semantic change (middle authorship, not corresponding).
- 2026Both papers land in the same year, concentrating the profile in machine learning for robotics: world-model planning on one side, VLA robustness evaluation on the other.
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.
- Name is common; the profile is name-resolved from public bibliographic data and keyed to ORCID 0000-0002-7467-2737; institution and country are not recorded.
- Bibliometrics come from the provided OpenAlex-derived snapshot only; citation counts, publication volume, and per-year activity are absent, so output, impact, and activity are unscored.
- Only two papers (both 2026 arXiv preprints) are reviewed; the reproducibility mean of 55 rests on just those two papers.
- Both papers are arXiv preprints; peer-reviewed venue acceptance is not recorded.