Justin Svegliato

University of California, Berkeley · United States

Unclaimed · AI-structured profileOpenAlex

AI researcher at UC Berkeley spanning multi-robot coordination and efficient long-form generation with language models

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—

No score yet: the underlying publication and coverage records needed for this axis are incomplete in the current snapshot. It would measure total scholarly output — cumulative works, venue spread, and productivity trends — which cannot be computed reliably from the available data.

Impact—

No score yet: citation-linked records are too sparse in the snapshot to rate this axis. It would measure scholarly influence — citations, citation trajectory over time, and uptake of the author's methods by other research groups — none of which can be estimated from the current data.

Focus0/100

Focus scores 0, reflecting an evenly split portfolio: the field distribution assigns 50% to robotics (multi-robot coordination / embodied AI) and 50% to human-computer interaction and NLP (efficient long-form generation). Topic counts confirm the spread — AI-based problem solving and planning (21), reinforcement learning in robotics (15), ethics and social impacts of AI (9), topic modeling (7), and ethics in business and education (6) — indicating deliberate breadth rather than a single niche.

Collaboration53/100

Collaboration scores 53, a moderate figure built on 13 distinct coauthors. The frequent-coauthor list splits along the two research lines: an NLP cluster (Zhengping Jiang, Gonzalo Ramos, Jina Suh, Benjamin Van Durme, Elias Stengel-Eskin, Sam Thomson) and a robotics cluster (Kyle H. Wray, Saaduddin Mahmud, Mason Nakamura). Both catalog papers place him in a middle authorship position without a corresponding-author role.

Activity75/100

Activity scores 75, the strongest scored dimension. Yearly counts show 1-2 works from 2013 to 2019, a ramp to 6 in 2020, a peak of 8 in 2021, 5 in each of 2022 and 2023, a dip to 3 in 2024, a rebound to 6 in 2025, and 3 indexed already in 2026. With 12 works over the last three years, the record indicates a sustained and currently active pace.

Reproducibility38/100

Reproducibility scores 38, the mean readiness rating across the 2 Evolve-reviewed papers. RoboTalk is rated low: the jump from ~2% to 77% held-out success for Qwen3-VL-8B comes with no significance tests and an evaluation that mirrors the training scaffold. FrankenReport is rated medium: its per-section quality predictions are a promising formulation, but headline gains are simulated and judge-based.

Research summary

Justin Svegliato is a US-based AI researcher at the University of California, Berkeley whose indexed record spans two equally weighted lines of work: robotics (multi-robot coordination / embodied AI) and human-computer interaction and NLP (efficient long-form generation), each carrying 50% of the field distribution in the bibliometric snapshot. Topic frequencies reinforce this breadth — 21 works tagged AI-based problem solving and planning, 15 in reinforcement learning in robotics, 9 in ethics and social impacts of AI, 7 in topic modeling, and 6 in ethics in business and education. Activity is the strongest scored dimension (75): 12 works in the last three years, with an indexed peak of 8 works in 2021, 5 in each of 2022 and 2023, a dip to 3 in 2024, and a rebound to 6 in 2025. The snapshot lists h-index 8 and i10-index 7, but the citation-based output and impact axes remain unscored for insufficient data. Two 2026 arXiv preprints are reviewed in the Evolve catalog — RoboTalk on multi-robot communication learned from multimodal demonstrations, and FrankenReport on early exiting in long-form generation via expected value of computation. Both list him as a middle author without the corresponding role; reproducibility readiness is rated low for RoboTalk and medium for FrankenReport, yielding a reproducibility mean of 38 across 2 reviewed papers.

Collaboration pattern

With 13 distinct coauthors and a moderate collaboration score of 53, Svegliato works in small-to-mid sized teams that track his two research lines. The frequent-coauthor list pairs an NLP cluster — Zhengping Jiang, Gonzalo Ramos, Jina Suh, Benjamin Van Durme, Elias Stengel-Eskin and Sam Thomson — with a robotics cluster including Kyle H. Wray, Saaduddin Mahmud and Mason Nakamura. In both cataloged papers he appears as a middle author rather than corresponding author, suggesting a contributor role.

Research keywords

multi-robot coordinationvision-language modelsdecentralized robot controlDec-POMDPssynthetic data generationsupervised fine-tuningearly exitinglong-form report generationdeep research agentsexpected value of computationLLM-as-a-judge evaluation

Highlights

RoboTalk: synthetic dual-robot demonstrations lift a VLM from ~2% to 77% held-out success
The Evolve review finds the synthetic dual-robot dataset credible for teaching Qwen3-VL-8B multi-robot communication and decentralized coordination from multimodal demonstrations, while flagging missing significance tests and an evaluation that mirrors the training scaffold (reproducibility: low).
FrankenReport frames early exiting in long-form generation as expected value of computation
The review credits a promising formulation that predicts per-section quality gains to decide when to stop computing in deep-research report generation, while noting the headline gains are simulated and rely on LLM-as-a-judge scoring (reproducibility: medium).
A shared theme: spending computation only where it pays off
Across both 2026 preprints, decentralized multi-robot coordination and early-exit report generation treat computation as a budget to be allocated — a throughline connecting the robotics and NLP halves of this researcher's profile.

Timeline

2020Indexed output ramps up to 6 works, after single-digit annual counts from 2013 onward.
2021Peak indexed year with 8 works.
2024Output dips to 3 works, the lowest annual count since 2019.
2025Rebound to 6 works; 12 works are recorded over the last three years.
2026Two Evolve-reviewed arXiv preprints appear — RoboTalk (multi-robot coordination) and FrankenReport (efficient long-form generation) — both with middle authorship; 3 works are indexed for the year so far.

Papers in the Evolve catalog

Frequent co-authors

Zhengping Jiang×1 Gonzalo Ramos×1 Jina Suh×1 Shiqian Rachel Ng×1 Elias Stengel-Eskin×1 Benjamin Van Durme×1 Andy Huntington×1 Sam Thomson×1 Dorian Benhamou Goldfajn×1 Mason Nakamura×1 Saaduddin Mahmud×1 Kyle H. Wray×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 derive from an OpenAlex-sourced snapshot; no ORCID is linked, so this profile is name-resolved from public bibliographic data.
Output and impact axes are unscored because the underlying citation and coverage data were insufficient in the snapshot.
The reproducibility score (38) averages readiness ratings from only 2 Evolve-reviewed papers; the wider record has not been reviewed.
The 2026 count is a partial year, and career details such as awards or roles fall outside the scope of this data.