Texas Tech researcher on vision-language-action model reliability and adversarial security of autonomous vehicles
2
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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: the snapshot contains no publication counts, venue history, or yearly output totals for this researcher, so overall research output cannot be computed. This axis would measure cumulative publication volume and productivity across venues and years.
Impact—
Not yet scored: no citation counts, h-index, or influence indicators are available in the provided data. This axis would measure how widely the work is cited and its influence within robot learning and autonomous-driving security communities.
Focus100/100
Scores a perfect 100 on focus: in the current snapshot the field distribution is confined to a single field, robot learning / vision-language-action models (1 cataloged paper), and the second cataloged paper on adversarial calibration attacks on autonomous vehicles sits in the adjacent autonomous-systems-safety space. The cataloged work consistently targets trustworthiness of learning-based autonomous systems, from VLA failure prediction to attack surfaces.
Collaboration32/100
Collaboration scores 32, reflecting a small, concentrated network: the snapshot records 4 distinct coauthors, with recurring partners Qingzhao Zhang, Kang G. Shin, Kaiwen Zhu, and Dongfang Liu. Authorship positions vary — first author on the autonomous-vehicle calibration-attack paper and last author on the Evolve-reviewed VLA-Scope paper — consistent with hands-on lead roles in some projects and senior roles in others on small teams.
Activity—
Not yet scored: beyond the two 2026 arXiv papers listed, no publication cadence, submission frequency, or yearly activity data is available. This axis would measure research tempo and continuity over time, such as outputs per year and recent momentum.
Reproducibility55/100
Reproducibility scores 55 (reproMean 55), computed from exactly one reviewed paper: VLA-Scope, rated medium. The Evolve review credits careful, ablation-rich experiments on OpenVLA/LIBERO, but notes the study covers a single model, is simulation-only, and reports no significance tests — factors that cap confidence in independent replication.
Research summary
Liangkai Liu is a researcher at Texas Tech University whose cataloged work targets the trustworthiness of learning-based autonomous systems from two complementary angles: predicting when vision-language-action (VLA) models are about to fail, and demonstrating how autonomous vehicles can be attacked. The Evolve-reviewed paper VLA-Scope: Shift-Aware Failure Prediction for Vision-Language-Action Models (arXiv, 2026), with Liu in the last-author position, presents a two-stage out-of-distribution and failure-risk predictor; the review credits a careful, ablation-rich OpenVLA/LIBERO study while flagging that it evaluates a single model, is simulation-only, and reports no significance tests. Reproducibility readiness for this paper is medium (reproducibility mean 55, computed from a single reviewed paper). In the same year Liu appears as first author of Adversarial Calibration Attack on Autonomous Vehicles, extending the agenda from runtime monitoring toward adversarial attack surfaces around vehicle calibration. The bibliometric snapshot is currently small: four distinct coauthors, repeated collaboration with Qingzhao Zhang, Kang G. Shin, Kaiwen Zhu, and Dongfang Liu, and a field distribution confined entirely to robot learning / vision-language-action models — yielding a maximal focus score of 100 and a modest collaboration score of 32. Output, impact, and activity axes remain unscored because the citation and volume data they require are not yet available; read this as an early-stage, name-resolved profile built from public bibliographic records (ORCID 0000-0002-6149-9859).
Collaboration pattern
The snapshot shows a compact collaboration footprint: 4 distinct coauthors and a collaboration score of 32. Recurring partners are Qingzhao Zhang, Kang G. Shin, Kaiwen Zhu, and Dongfang Liu. Liu takes different seats across papers — first author on the 2026 autonomous-vehicle calibration-attack study and last author on the Evolve-reviewed VLA-Scope paper — a pattern consistent with hands-on lead work in some projects and senior or supervisory roles in others within a small team.
Research keywords
Highlights
Two-stage shift-aware failure prediction for VLA models
First-author adversarial calibration attack on autonomous vehicles
In 2026 Liu appears as first author of Adversarial Calibration Attack on Autonomous Vehicles, extending the research agenda from monitoring VLA reliability toward offensive security analysis of calibration in autonomous driving. The paper is in the catalog but has no Evolve review, keywords, or slug yet.
A dual track toward trustworthy autonomous systems
Taken together, the two 2026 papers — one first-author, one last-author — sketch a coherent program: predicting failures before they happen (VLA-Scope) and proactively exposing vulnerabilities (calibration attacks), all within autonomous-systems safety, where the snapshot's field distribution is fully concentrated (focus score 100).
Timeline
- 2026Published VLA-Scope: Shift-Aware Failure Prediction for Vision-Language-Action Models on arXiv as last author; Evolve reviewed it and rated reproducibility readiness medium.
- 2026First author of Adversarial Calibration Attack on Autonomous Vehicles (arXiv), adding adversarial security of autonomous vehicles to the research agenda.
- 2026Snapshot bibliometrics took shape: 4 distinct coauthors, single-field distribution in robot learning / vision-language-action models (focus score 100), and a reproducibility mean of 55 from one reviewed paper.
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.
- Bibliometric scores are computed from a small OpenAlex-based snapshot covering only two 2026 arXiv papers and 4 distinct coauthors; they may not reflect the full publication record.
- Output, impact, and activity axes are unscored because citation counts, publication totals, and yearly cadence data were not available in the input data.
- The reproducibility mean (55) is based on a single reviewed paper (VLA-Scope) and should be read as an early signal, not a stable estimate.
- Profile is name-resolved from public bibliographic data linked to ORCID 0000-0002-6149-9859, with affiliation listed as Texas Tech University; the second cataloged paper has no Evolve review, keywords, or slug yet.