Early-career researcher evaluating LLM reliability, agentic social reasoning, and open robot policy benchmarks
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-08) 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 verified publication-count, venue, or output-history data is available for this record yet, so output cannot be scored. This axis would measure total scholarly output, venue quality, and publication consistency over time.
Impact—
Citation and h-index data are absent from this snapshot, so impact is unscored. This axis would measure citation-based influence — total citations, citation velocity, and uptake of the 2026 preprints once indexed.
Focus0/100
Insufficient topic-classified publication data exists to score research focus. This axis would measure thematic concentration versus dispersion across an author's output, derived from field and keyword distributions over time.
Collaboration60/100
The only scored dimension: 55/100, computed from a record showing 15 distinct coauthors, 12 of whom recur across papers — including Jianzhe Lin, Xiaolin Li, Jubin Chheda, and Robert Douglas. All three cataloged papers are middle-author collaborations with no corresponding-author record, indicating steady team-based work in group settings rather than solo or leading-author output; the mid-range score reflects moderate collaborator breadth to date.
Activity—
Recent-productivity data is insufficient: the visible record covers only three 2026 preprints without a longer publication history. This axis would measure publication cadence and output continuity across recent years.
Reproducibility58/100
No reproducibility evidence — code releases, artifact links, or Evolve reproReadiness ratings — is available for the cataloged papers, so this axis is unscored. It would measure how reproducible and artifact-ready the author's outputs are.
Research summary
Fei Wang is an early-stage researcher whose cataloged work centers on evaluating large language models (LLMs) and their embodied and agentic extensions. The Evolve catalog lists three 2026 arXiv preprints, each with Fei Wang in a middle author position and none as corresponding author: IndustrialVLA-Bench, a traceable multi-axis benchmark for evaluating open robot policy models; “Clarification Is Not Correction: LLMs Fail to Let Go”, a study of how LLMs mishandle user clarifications in dialogue; and ReAdapt, which targets relational social reasoning failures in LLM agents. Together they span robotics evaluation, conversational robustness, and agent social intelligence, suggesting a profile oriented toward measurement and behavioral analysis of modern AI systems rather than a single architectural contribution. The bibliometric record is still thin: the radar scores only collaboration (55/100, derived from 15 distinct coauthors), while output, impact, focus, activity, and reproducibility lack sufficient data to score. Twelve recurring collaborators, including Jianzhe Lin, Xiaolin Li, and Jubin Chheda, indicate stable team-based work. No institution, country, or ORCID is attached, and because “Fei Wang” is a common name, this profile is name-resolved from public bibliographic data and should be read as provisional. All cataloged entries are recent preprints, so citation and venue signals have not yet accumulated.
Collaboration pattern
Fei Wang works in teams: all three cataloged papers are middle-author efforts with no corresponding-author record. The bibliometric record shows 15 distinct coauthors and 12 recurring collaborators, including Jianzhe Lin, Xiaolin Li, Jubin Chheda, Robert Douglas, Rajeshkumar Golani, and Yiqi Wang, spanning what appear to be robotics-evaluation and LLM-behavior project clusters. The collaboration score of 55/100 sits mid-range, consistent with steady group participation rather than an established independent or leading-author network.
Research keywords
Highlights
IndustrialVLA-Bench: traceable, multi-axis evaluation of open robot policies
A 2026 arXiv preprint introducing a benchmark that evaluates open robot policy models across multiple axes with traceability, positioning rigorous evaluation as central to embodied AI progress. Fei Wang is a middle author on the paper.
Clarification Is Not Correction: diagnosing LLM stubbornness in dialogue
This 2026 study shows that LLMs fail to “let go” of earlier answers after a user clarification, documenting a behavioral failure mode directly relevant to conversational reliability. Fei Wang is a middle author on the paper.
ReAdapt: giving LLM agents relational social reasoning
A 2026 preprint, “When LLM Agents Fail to Read the Room”, proposes ReAdapt to address relational social reasoning failures in LLM agents, extending the evaluation agenda from task competence to social competence. Fei Wang is a middle author on the paper.
Timeline
- 2026Co-authored IndustrialVLA-Bench, a traceable multi-axis benchmark for evaluating open robot policy models (arXiv preprint, middle author).
- 2026Co-authored “Clarification Is Not Correction: LLMs Fail to Let Go”, examining how LLMs mishandle user clarifications in dialogue (arXiv preprint, middle author).
- 2026Co-authored ReAdapt, targeting relational social reasoning failures in LLM agents (arXiv preprint, middle author).
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.
- “Fei Wang” is a common name; this profile is name-resolved from public bibliographic data and may not correspond to a single individual.
- Bibliometric scores come from an OpenAlex-derived snapshot; only the collaboration axis has enough data to score, so the other radar axes are deliberately left unscored.
- No institution, country, or ORCID is attached to this record, so affiliation-level claims are absent by design.
- All three cataloged works are 2026 arXiv preprints with no venue, citation, or reproducibility metadata yet, so impact-related statements are limited.