UC Irvine researcher spanning robot learning and NLP; 15 h-index; VLAs for dynamic manipulation, LLM text autoencoders.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
15
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—
Not yet scored: the snapshot lacks complete publication-count coverage, so total volume, venue mix, and authorship-position share cannot be computed reliably. This axis would measure annual output, venue tiers, and first/corresponding-author share across the full record.
Impact—
Not yet scored: citation coverage is too incomplete for a reliable rating, though the snapshot lists an h-index of 15 and an i10-index of 20. This axis would measure citation volume, trajectory, and influence relative to field norms.
Focus0/100
Focus scores 0 because the field distribution is an even 50/50 split between robot learning/embodied AI and machine learning/NLP, leaving no dominant concentration. Topic counts confirm the breadth: reinforcement learning in robotics (30), robot manipulation and learning (24), and multimodal machine learning applications (23) sit alongside smaller clusters in topic modeling (6) and domain adaptation/few-shot learning (6). The record spans two active subfields in parallel rather than one niche.
Collaboration39/100
Collaboration scores 39, resting on a compact recorded network: the snapshot lists only 6 distinct co-authors. The two 2026 reviewed papers place the author in middle and last positions rather than solo work, and recurring partners — Erik B. Sudderth, Alexander C. Berg, Zhenyang Feng, Jimin Heo, Arkanath Pathak — point to small, stable teams rather than broad cross-institution collaboration.
Activity83/100
Activity scores 83, the strongest dimension on file. The record shows 16 works in the recent three-year window and annual counts of 3-7 papers every year from 2017 through 2025, rising to 10 works in 2026 — the highest annual figure recorded. Both Evolve-reviewed papers are 2026 arXiv releases, consistent with an actively running research program.
Reproducibility38/100
Reproducibility averages 38 over the 2 papers Evolve has reviewed. The LLMAE text-autoencoder paper is rated medium readiness, with the verdict flagging missing baselines and unreported variance gaps despite strong results; TEMPO is rated low, noting few baselines and no error bars even with strong real-robot gains. Larger baseline tables and statistical reporting would move this score.
Research summary
Unnat Jain is a researcher at the University of California, Irvine whose published record splits evenly between robot learning/embodied AI and machine learning/NLP (50% each in the field distribution). The bibliometric snapshot shows an h-index of 15, an i10-index of 20, and 16 works in the recent three-year window, with annual output running 3-7 papers from 2017 to 2025 before reaching 10 works in 2026. Topic analysis concentrates on reinforcement learning in robotics (30 works), robot manipulation and learning (24), and multimodal machine learning applications (23), with smaller clusters in topic modeling (6) and domain adaptation/few-shot learning (6). Evolve has reviewed two 2026 arXiv papers: TEMPO, which extends vision-language-action models with motion tokens and proprioceptive history for dynamic robot manipulation (last-author position), and a paper repurposing a 270M-parameter pre-trained LLM as a high-fidelity continuous text autoencoder via masked bottlenecks (middle-author position). Reproducibility readiness averages 38 across these reviews, with both verdicts citing missing baselines or unreported variance. The recorded collaboration network is compact at 6 distinct co-authors, with recurring partners including Erik B. Sudderth and Alexander C. Berg. The profile is built from OpenAlex-derived bibliometrics plus Evolve's reviews; output and impact axes remain unscored.
Collaboration pattern
The recorded network is compact: only 6 distinct co-authors appear in the bibliometric snapshot, and the collaboration dimension scores 39. Recurring partners include Erik B. Sudderth, Alexander C. Berg, Zhenyang Feng, Jimin Heo, and Arkanath Pathak. On the two 2026 reviewed papers Jain sits in middle and last author positions, consistent with senior-authorship roles inside small, stable teams rather than large multi-institution consortia.
Research keywords
Highlights
TEMPO: motion tokens and proprioceptive history for dynamic manipulation
A 270M pre-trained LLM repurposed as a near-lossless text autoencoder
2026 output peak: 10 works on record
The publication counts jump to 10 works in 2026 — the highest annual figure in a record that ran 3-7 per year from 2017 to 2025 — and the year includes both Evolve-reviewed papers.
Timeline
- 2017Earliest year in the publication counts, with 3 works on record.
- 2021Peak annual output of the 2017-2025 stretch, with 7 works.
- 20235 works, keeping the steady 3-7 annual cadence.
- 20253 works in the last complete year before the 2026 jump.
- 202610 works on record — the highest annual count — including both reviewed papers: the LLMAE text autoencoder and TEMPO for dynamic manipulation.
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 indicators (h-index, i10-index, counts by year, topics, co-authors) come from an OpenAlex-based snapshot; no ORCID is linked, so identity is name-resolved from public bibliographic data.
- The reproducibility mean of 38 rests on only 2 Evolve-reviewed papers and should be read as a small-sample signal.
- Output and impact dimensions are unscored because underlying coverage is incomplete; all quoted figures are taken as provided and not independently verified.
- The 2026 annual count (10 works) covers a year still in progress and may change.