First-author researcher on post-training quantization for vision-language-action models and robot policy deployment.
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-14) 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—
Not yet scored: only the two 2026 arXiv preprints in Evolve's catalog are available. This axis would measure total publication volume, venue placement, and author-role distribution over time.
Impact—
Not yet scored: no citation counts, h-index, or downstream-adoption signals are provided. This axis would measure how widely the quantization benchmarks and predictors are cited and reused by other researchers.
Focus0/100
The radar records focus at 0. The raw field distribution splits evenly — 0.5 and 0.5 — between two nearly identical labels, 'robot learning and efficient model deployment (quantization)' and 'robot learning and efficient inference (model quantization)'. Both cataloged papers therefore target one tightly defined problem, quantization for vision-language-action and world action models, so topical concentration is extremely narrow in practice; the nominal 0 score reflects a two-paper sample rather than scattered interests.
Collaboration44/100
Collaboration scores 44 on the radar. The two cataloged papers involve 8 distinct co-authors, including Meida Chen, Jinjia Guo, Jing Du, Yangming Shi, Qing Jin, Song Wang, and Yang Sui — a moderate-sized, recurring circle. Xu is first author on both papers but not corresponding author, indicating lead technical contributions within a supervised team rather than an independent research group so far.
Activity—
Not yet scored: the snapshot contains no publication-frequency or recency series beyond the two 2026 papers. This axis would measure annual publishing cadence and whether recent output is accelerating or slowing.
Reproducibility78/100
Reproducibility averages 78 across the 2 papers assessed. VLAQuantBench is rated high: its closed-loop protocol spans 409 runs and 94,574 episodes with transparent, recipe-level evaluation. Predict Before You Deploy is rated medium, with tiny calibration sets flagged as the scope limitation. The pattern suggests strong methodological transparency and evaluation pipelines others can rerun — the strongest dimension currently on record.
Research summary
Jiuyi Xu is an early-career researcher at Colorado School of Mines whose work sits at the intersection of robot learning and efficient model deployment. The Evolve catalog holds two first-author arXiv preprints, both dated 2026, forming a coherent agenda around post-training quantization (PTQ) for vision-language-action and world action models. The first, VLAQuantBench, delivers a closed-loop evaluation of PTQ across 409 runs and 94,574 episodes and shows that quantization effects are recipe-dependent; Evolve rates it high for reproducibility. The second, Predict Before You Deploy, predicts quantization-induced task degradation offline through calibrated action-deviation thresholds, enabling selective prediction and deferral; it is rated medium, with tiny calibration sets as the noted limitation. On the radar, reproducibility averages 78 across the two papers — the strongest recorded dimension — while collaboration scores 44 with 8 distinct co-authors, including Meida Chen, Jinjia Guo, Jing Du, Yangming Shi, Qing Jin, Song Wang, and Yang Sui. Focus records 0, though the raw field distribution splits evenly between two near-identical quantization-for-robot-learning labels, indicating extreme topical concentration in practice. Output, impact, and activity are not yet scored. Overall, this is an early-stage but sharply defined profile: a first-author contributor building evaluation infrastructure and deployment-risk tools for compressed robot policies.
Collaboration pattern
Jiuyi Xu works within a moderate-sized team: the two cataloged papers involve 8 distinct co-authors, with Meida Chen, Jinjia Guo, Jing Du, Yangming Shi, Qing Jin, Song Wang, and Yang Sui recurring most often. As first author on both papers but not corresponding author, Xu leads the technical core while senior collaborators anchor supervision — a pattern consistent with an early-career researcher embedded in a robotics and model-efficiency group.
Research keywords
Highlights
VLAQuantBench: a closed-loop yardstick for quantized vision-language-action models
Predicting quantization damage before deployment
Timeline
- 2026Publishes VLAQuantBench, a first-author closed-loop evaluation of post-training quantization for vision-language-action models, spanning 409 runs and 94,574 episodes.
- 2026Publishes Predict Before You Deploy, introducing offline prediction of quantization-induced task degradation for world action models via calibrated action-deviation thresholds.
- 2026Both cataloged works appear on arXiv in the same year as first-author (non-corresponding) contributions, establishing an early-stage, quantization-centered agenda at Colorado School of Mines.
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.
- Radar dimensions are computed from the bibliometric snapshot; output, impact, and activity lack sufficient data and are not scored.
- Coverage is limited to the two 2026 arXiv preprints in Evolve's catalog; citation counts, venue history, and career facts are not included.
- No ORCID or country is provided; the profile is name-resolved from public bibliographic data and disambiguation is not independently verified.
- Both papers are first-author preprints; corresponding-author status and peer-reviewed venue placement are not recorded.