Xingjie Fan

Tsinghua University · China

Unclaimed · AI-structured profileOpenAlex

Early-career researcher at Tsinghua University working on vision-language-action models and precise robot manipulation.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-16) 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—

Insufficient data: this axis would measure total publication volume, venue quality, and authorship role over time, but the snapshot contains only two 2026 arXiv preprints, so no reliable output score can be computed yet.

Impact—

Insufficient data: impact would measure citations, h-index, and downstream influence, but both works are 2026 preprints and the snapshot records an h-index and i10-index of zero, so no meaningful impact score exists yet.

Focus0/100

This axis scores 0, the lowest concentration band, reflecting a field distribution split exactly 50/50 between 'robotics / robot learning' and 'robot learning (vision-language-action models)' and three topics with two papers each. In practice the split is narrower than the score suggests: both cataloged papers share robot manipulation as a keyword, so the work spans two adjacent sub-fields rather than genuinely disconnected areas.

Collaboration66/100

Scoring 66, this is the profile's strongest supported signal: the two papers were written with 26 distinct coauthors, and twelve collaborators — including Tianyu Xin, Lijie Wang, and Zheng Lu — recur across the record. Both papers list Xingjie Fan in a middle authorship position without corresponding-author status, consistent with contribution inside a larger, stable team rather than solo or lead-author work.

Activity32/100

Scoring 32, activity reflects a very recent and thin record: countsByYear lists a single entry, 2026, with two works, and recent3yWorks equals 2 — both cataloged papers. There is no prior-year publication activity in the snapshot, so the score measures career stage and indexing lag as much as research tempo; it should rise if output continues beyond these two preprints.

Reproducibility55/100

The reproducibility score of 55 is computed across two papers (reproPapers: 2, reproMean: 55), both rated medium reproducibility readiness by Evolve. 'Acting in Meters' is described as well-ablated but weakened by single-seed statistics and no code release; 'LayerRoute' reports multi-benchmark gains up to +7.2 points on LIBERO-Long, but its small margins lack variance statistics. Code release and seed-level reporting would lift this axis.

Research summary

Xingjie Fan is an early-stage researcher at Tsinghua University whose public bibliographic record, as indexed by OpenAlex, currently consists of two 2026 arXiv preprints in robot learning. Both papers sit in the middle of the author list and address precision in vision-language-action (VLA) systems: 'Acting in Meters' introduces metric-interaction adapters (ICT and MAIF) that improve manipulation success for VLA and world-action models across simulation and real hardware, while 'LayerRoute' adds action-conditioned mixture-of-layers routing with action-state rereading, reporting gains of up to +7.2 percentage points on LIBERO-Long. Topic classifications group the work under reinforcement learning in robotics, robot manipulation and learning, and multimodal machine learning applications, and the field distribution splits evenly (0.5/0.5) between robotics/robot learning and VLA research. The snapshot records 26 distinct coauthors, with twelve recurring collaborators, giving a collaboration score of 66; activity is rated 32 because both works date to 2026, and the recorded h-index and i10-index are zero, consistent with a record too new for citations to accumulate. Evolve rates both papers at medium reproducibility readiness (mean 55), noting well-run ablations but single-seed statistics, missing variance reporting, and no code release for one paper. This is an early-stage profile expected to mature as the record grows.

Collaboration pattern

The record is strongly team-based: just two papers but 26 distinct coauthors, with twelve recurring collaborators including Tianyu Xin, Lijie Wang, Zheng Lu, and Yiming Wang. Xingjie Fan holds a middle authorship position on both papers and is not the corresponding author, suggesting a contributing role within a larger Tsinghua-affiliated group working on vision-language-action models and manipulation. The collaboration score of 66 reflects breadth of contacts relative to output.

Research keywords

vision-language-action modelsrobot manipulationrobot learningworld-action modelsmixture-of-layers routingrepresentation accessflow matchingpoint cloud geometrymetric scalereinforcement learning in roboticsmultimodal machine learningLIBERO benchmark

Highlights

Metric-interaction adapters for precise robotic manipulation
The 'Acting in Meters' preprint introduces ICT and MAIF adapters that consistently improve VLA and world-action model manipulation success in both simulation and real hardware; ablations are thorough, but single-seed statistics and no code release temper certainty.
Action-conditioned mixture-of-layers routing for VLA policies
The 'LayerRoute' preprint adds action-conditioned VLM-layer routing and action-state rereading to vision-language-action policies, delivering multi-benchmark gains of up to +7.2 percentage points on LIBERO-Long, though small margins lack variance statistics.
A consistent thread: precision in VLA-based manipulation
Across both 2026 preprints, the recurring theme is making VLA-driven manipulation more precise and reliable — at the interaction level through metric-scale geometry, and at the representation level through action-conditioned layer access.

Timeline

2026First cataloged preprint: 'Acting in Meters' introduces metric-interaction adapters (ICT + MAIF) for precise robotic manipulation.
2026'LayerRoute' proposes action-conditioned mixture-of-layers routing, reporting gains up to +7.2 percentage points on LIBERO-Long.
2026The OpenAlex snapshot records 2 works in 2026 with 26 distinct coauthors, marking the start of this researcher's indexed record.

Papers in the Evolve catalog

Frequent co-authors

Tianyu Xin×2 Lijie Wang×1 Zheng Lu×1 Yiming Wang×1 Heyang Yu×1 Kenghou Hoi×1 Bowen Hu×1 Di Cui×1 Haoran Liao×1 Wanqi Zhong×1 Yizhao Xu×1 Ziliang Wang×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 derived from OpenAlex only; no ORCID is linked, so identity is name-resolved from public bibliographic data.
The record contains only two 2026 arXiv preprints; h-index and i10-index are zero, and the output and impact axes are unscored.
Only papers reviewed by Evolve are covered; any other works fall outside this profile's scope.
Both papers are rated medium reproducibility readiness; one lacks a code release, and reported gains lack variance statistics.