Yifan Wang

China People's Public Security University · United States

Unclaimed · AI-structured profileOpenAlex

Early-stage researcher bridging embodied-AI (VLA) engineering and NLP-based content moderation & web-agent safety

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

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—

The output axis is not yet scored: beyond the 6 works captured for 2026, this OpenAlex identity has no verified venue or publication-volume record. Output would measure total publications and their venue quality, and is left unscored until the record consolidates.

Impact—

Impact data is insufficient: the snapshot reports h-index 0 and i10-index 0, meaning citations are not yet linked to this identity. Impact would measure citation influence and uptake of the work, and is deliberately left unscored here.

Focus0/100

Focus scores 0: the field distribution is an exact 50/50 split between robotics/robot learning (embodied AI systems) and NLP for content moderation and web-agent safety, and the five recorded topics — AI-based problem solving and planning (2), complex network analysis, nonlinear PDEs, reinforcement learning in robotics, and robot manipulation — show breadth rather than a single concentrated line.

Collaboration72/100

Collaboration scores 72. The OpenAlex-linked record lists 36 distinct coauthors, of which 12 recur — including Yinhao Li, Weixin Mao, and Zihan Lan — suggesting a stable team core with a continuously extending outer circle. Both reviewed 2026 preprints are middle-position, non-corresponding contributions, consistent with team-based engineering and benchmark-building rather than solo work.

Activity57/100

Activity scores 57, reflecting a short but dense record: countsByYear attributes all 6 works to 2026, and recent3yWorks is likewise 6, so the entire captured output falls within a single year. With no earlier years recorded, this reads as an early-stage activity profile rather than a demonstrated multi-year cadence; the score should firm up as later windows accumulate.

Reproducibility55/100

Reproducibility scores 55, the mean across 2 papers with repro-readiness assessments. Both reviewed papers rate medium: FluxVLA Engine offers real integration and deployment evidence but all quantitative results are project-reported and statistically uncontrolled, while RiskChainBench is well designed yet constrained by synthetic data and a single-run evaluation, limiting generality.

Research summary

Yifan Wang (China People's Public Security University) is an early-stage researcher whose OpenAlex-linked profile shows a compact but fast-moving 2026 record: six works in the snapshot year, all falling inside the recent three-year window. The profile is deliberately two-sided — field distribution splits exactly 50/50 between robotics/robot learning (embodied AI systems) and natural language processing for content moderation and web-agent safety. Two arXiv preprints reviewed by Evolve anchor this split. The first, FluxVLA Engine, is a one-stop engineering platform for vision-language-action models, covering data-to-deployment workflows, real-time and action chunking, and CUDA Graph inference acceleration; Evolve credits it with unusually candid, real integration evidence while cautioning that quantitative results are project-reported and statistically uncontrolled. The second, RiskChainBench, is a two-stage benchmark connecting obfuscated platform message restoration — including Chinese homophone evasion — to evidence-grounded web investigation with VLM agents, rated well designed but limited by synthetic data and single-run evaluation. The radar reads collaboration 72 (36 distinct coauthors, 12 recurring) and reproducibility 55 (mean across two medium-readiness papers), with activity at 57 and focus at 0, reflecting the even two-field split. Output and impact are not scored: citation metrics (h-index and i10-index both 0) are not yet linked to this identity. As 'Yifan Wang' is a common name, this profile is name-resolved from public bibliographic data.

Collaboration pattern

The collaboration footprint is team-based: 36 distinct coauthors appear in the snapshot, of which 12 recur — Yinhao Li, Weixin Mao, Zihan Lan, Jikun Rong, Qirui Hu, Yiming Zhang, Weipeng Deng, Bowen Shen, Minzhao Zhu, Yiming Mao, Yan Yang, and Chenguang Cui. Both reviewed papers are middle-position, non-corresponding contributions, suggesting this author operates within a larger engineering or benchmark team rather than as a solitary lead.

Research keywords

vision-language-action modelsworld-action modelsrobot learning infrastructureaction chunkingreal-time chunking (RTC)CUDA graph inference accelerationcontent moderationobfuscated text restorationChinese homophone evasionweb agentsVLM agents

Highlights

FluxVLA Engine: one-stop VLA engineering platform
Evolve's review credits this embodied-intelligence platform with unifying data-to-deployment workflows and unusually candid integration evidence — real-time/action chunking and CUDA Graph inference acceleration included — while flagging that all quantitative results are project-reported and statistically uncontrolled.
RiskChainBench: from obfuscated messages to web-agent risk
A two-stage benchmark linking obfuscated platform message restoration — including Chinese homophone evasion — to evidence-grounded web investigation with VLM agents; well designed, though synthetic data and single-run evaluation limit generality.
A two-field 2026 output burst
All 6 works captured in the 2026 snapshot split evenly between embodied-AI/robot-learning infrastructure and NLP content-moderation safety, with recorded topics spanning AI planning, complex network analysis, nonlinear PDEs, and robot manipulation.

Timeline

2026First captured output year: OpenAlex attributes all 6 recorded works to 2026, with recent3yWorks likewise at 6.
2026FluxVLA Engine, a one-stop VLA engineering platform for embodied intelligence, appears on arXiv and is reviewed by Evolve as candid but statistically uncontrolled.
2026RiskChainBench, a benchmark tying obfuscated message restoration to evidence-grounded web investigation, appears on arXiv and is reviewed by Evolve as well designed but limited by synthetic data.

Papers in the Evolve catalog

Frequent co-authors

Yinhao Li×1 Weixin Mao×1 Zihan Lan×1 Jikun Rong×1 Qirui Hu×1 Yiming Zhang×1 Weipeng Deng×1 Bowen Shen×1 Minzhao Zhu×1 Yiming Mao×1 Yan Yang×1 Chenguang Cui×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.

Bibliometrics derive from an OpenAlex-linked snapshot only; output and impact are unscored, and h-index/i10-index read 0 because citations are not yet linked to this identity.
'Yifan Wang' is a common name; this profile is name-resolved from public bibliographic data and may not capture all of this person's work.
The source country field reads 'US' while the listed institution is China People's Public Security University — an apparent metadata inconsistency retained as provided.
Both reviewed papers are 2026 arXiv preprints whose quantitative results are project-reported and statistically uncontrolled; no peer-reviewed venue record appears in the snapshot.