Low-bit quantization and CPU-level speedups for vision-language-action models in robot manipulation
2
papers here
2
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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 publication-volume data is available in this snapshot, so the output axis is unscored. It would measure total and per-year publication counts across the researcher's full indexed record, beyond the two arXiv papers Evolve has reviewed.
Impact—
Citation-based impact metrics are not available here, so this axis is unscored. It would measure citations, citation velocity, and uptake of the researcher's work across the literature — none of which the current record provides.
Focus0/100
The raw field distribution is an even 0.5/0.5 split between two adjacent labels — 'robot learning / efficient inference (VLA model quantization)' and 'robot learning (VLA models for robotic manipulation)' — which yields the radar's focus score of 0 because no single label dominates. Substantively, both cataloged 2026 papers attack the same problem, efficient inference for VLA manipulation models, so topical concentration is tight despite the even label split.
Collaboration44/100
The radar scores collaboration at 44 from 8 distinct coauthors in the cataloged record. The recurring set — An T. Le, Hung T. Ho, Quang D. Nguyen, Thanh Q. Duong, Ngan Le, Meng Guo, Vien A. Ngo, and Hoang M. Truong — suggests a stable, compact author group. Nguyen led vla.simd as first author and joined FoldQuantVLA as a middle author, showing both leading and supporting roles within a small team.
Activity—
Publication-timing data beyond the two 2026 arXiv papers is absent, so the activity axis is unscored. It would measure publishing cadence and year-over-year consistency, which requires a longer indexed history than this snapshot provides.
Reproducibility55/100
The radar scores reproducibility at 55, a mean over two assessed papers (reproMean 55, reproPapers 2), both rated medium repro-readiness. Evolve's reviews credit rigorous timing analysis in vla.simd and honest statistics in FoldQuantVLA, while cautioning that simulation success trails top baselines and that headline robot gains rest on small, confounded trials.
Research summary
Khanh D. Nguyen works on making vision-language-action (VLA) models practical for real robotic manipulation, with the cataloged record concentrated on efficient inference. In 2026, Nguyen led vla.simd as first author: a CPU-only inference engine for language-conditioned manipulation combining SIMD micro-kernels with INT8 quantization and action chunking. Evolve's review credits roughly 1.4x median speedup and an IMPACT policy sustaining 30 actions per second on a Raspberry Pi 5, backed by rigorous timing analysis, while noting simulation success trails the strongest baselines. Also in 2026, Nguyen contributed as a middle author to FoldQuantVLA, a consistent-folding W4A4 post-training quantization scheme using Hadamard transforms and GPTQ, deployed natively through TensorRT INT4 plugins; Evolve judged the speedups credible and the statistics honest, but flagged that headline robot gains rest on small, confounded trials. Bibliometrically, the snapshot records 8 distinct coauthors and an even 0.5/0.5 split across two adjacent topic labels, so the radar's focus score reads 0 despite substantively tight concentration. Reproducibility scores 55, averaged over two assessed papers both rated medium readiness. Output, impact, and activity axes lack sufficient data and remain unscored; no institution, country, or ORCID is attached, so the profile is name-resolved from public bibliographic data.
Collaboration pattern
Nguyen's cataloged record involves 8 distinct coauthors, and the radar's collaboration score is 44. The recurring group — An T. Le, Hung T. Ho, Quang D. Nguyen, Thanh Q. Duong, Ngan Le, Meng Guo, Vien A. Ngo, Hoang M. Truong — recurs across the record. Roles alternate: first author on vla.simd, middle author on FoldQuantVLA, consistent with a small, stable team sharing hardware, quantization, and robotics evaluation workloads.
Research keywords
Highlights
First-author CPU inference engine for VLA manipulation
Consistent-fold W4A4 quantization deployed natively in TensorRT
Timeline
- 2026Published vla.simd on arXiv as first author: a CPU inference engine for language-conditioned manipulation with SIMD micro-kernels and INT8 quantization.
- 2026Co-authored FoldQuantVLA (middle position): native low-bit (W4A4) quantization of vision-language-action models with TensorRT INT4 deployment.
- 2026Both cataloged papers land in the same year, forming a concentrated line on efficient inference for VLA models.
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 come from a limited OpenAlex-based snapshot; the output, impact, and activity axes lack sufficient data and are unscored.
- Coverage is limited to two 2026 arXiv papers reviewed by Evolve; citation counts, h-index, and peer-reviewed venue status are outside this profile's scope.
- No institution, country, or ORCID is attached; this profile is name-resolved from public bibliographic data, and 'Khanh D. Nguyen' may be a common name, so identity is not guaranteed.
- The reproducibility score (55) rests on only two assessed papers.