Gianluca Bardaro

Politecnico di Milano · Italy

Unclaimed · AI-structured profileORCID OpenAlex

Politecnico di Milano researcher in robot learning, efficient VLA inference, and pedestrian trajectory prediction.

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—

Publication-output scoring is not yet available: the current OpenAlex record is too sparse or fragmented to compute a reliable per-year output measure. This axis would normally count works per year and their venue types.

Impact—

Impact scoring is not yet available: citation-linked aggregates for this OpenAlex identity are incomplete, so no reliable impact score can be reported. This axis would measure citation counts, h-index-based influence, and uptake of the reviewed papers.

Focus0/100

Focus scores 0/100, indicating output evenly split rather than concentrated: the field distribution assigns 0.5 to robotics/robot learning and 0.5 to computer vision for pedestrian trajectory prediction. Topic counts reinforce this breadth — 'Robotics and Automated Systems' (11 works) and 'Robot Manipulation and Learning' (8) sit beside 'Model-Driven Software Engineering Techniques' (7) and 'Advanced Software Engineering Methodologies' (6), showing a dual robotics-and-software-engineering profile.

Collaboration46/100

Collaboration scores 46/100, a moderate level: the recent three-year window covers 12 works with 9 distinct coauthors, and both catalog papers are middle-authored rather than solo or corresponding. Recurring partners include Riccardo Andrea Izzo, Subramanian Ramamoorthy, Matteo Matteucci, Alessandro Suglia, and Simone Mentasti, suggesting small team-based projects across institutions.

Activity75/100

Activity scores 75/100, reflecting a clearly active recent phase: 12 works in the last three years, with per-year counts rising from 1 (2024) to 4 (2025) and 7 already recorded in 2026. Earlier records date back to 2014, indicating a sustained but uneven publication rhythm over more than a decade.

Reproducibility78/100

Reproducibility scores 78/100, the average Evolve rating across 2 reviewed papers. The flow-matching VLA early-exit paper rates high, while MamMA rates medium; reviewers noted strong simulated gains but configurations tuned on evaluation data and no real-robot tests in the former, and narrow baselines in the latter.

Research summary

Gianluca Bardaro is a researcher at Politecnico di Milano whose profile is name-resolved from an OpenAlex record tied to ORCID 0000-0002-6695-0012 and spans robotics, robot learning, and vision-based prediction. The topic distribution is led by 'Robotics and Automated Systems' (11 works), 'Robot Manipulation and Learning' (8), and 'Reinforcement Learning in Robotics' (6), alongside a visible software-engineering strand ('Model-Driven Software Engineering Techniques', 7; 'Advanced Software Engineering Methodologies', 6). The field split is even — 0.5 robotics/robot learning and 0.5 pedestrian trajectory prediction — which is why the focus dimension scores 0. Recent output is clearly active: 12 works in the last three years, rising from 1 (2024) to 4 (2025) and 7 already recorded in 2026, for an activity score of 75. Two 2026 arXiv preprints are cataloged, both middle-authored: decoupled early exits for task-dependent compute allocation in flow-matching vision-language-action models (high reproducibility, though configurations were tuned on evaluation data and no real-robot tests are reported), and MamMA, a Mamba-based pedestrian trajectory predictor using LiDAR occupancy maps and awareness states (medium reproducibility; narrow baselines). The average reproducibility rating across these 2 reviewed papers is 78. Collaboration is moderate (46): 9 distinct coauthors, including recurring partners Riccardo Andrea Izzo, Subramanian Ramamoorthy, and Matteo Matteucci. Output and impact scores are not yet computable from the current record.

Collaboration pattern

Works in small co-authored teams rather than solo: 9 distinct coauthors across 12 works in the last three years, and both catalog papers are middle-authored, non-corresponding contributions. The recurring partner list includes Matteo Matteucci, Simone Mentasti, Riccardo Andrea Izzo, Rimvydas Rubavicius, Subramanian Ramamoorthy, Alessandro Suglia, Juncen Long, and Xiaofeng Jin, consistent with project-based robotics collaborations across institutions.

Research keywords

vision-language-action modelsflow matchingearly-exit inferenceKV cacheconditional computationinference efficiencypedestrian trajectory predictionstate space modelsoccupancy grid mapsLiDAR perceptionreinforcement learning in roboticsrobot manipulation and learning

Highlights

Decoupled early exits for flow-matching VLAs
Evolve rates this 2026 arXiv work high for reproducibility: decoupled early exits across the vision-language model, expert depth, and denoising steps, combined with KV-cache synthesis, allocate compute per task and deliver strong gains in simulation — though configurations were tuned on evaluation data and no real-robot tests are reported.
MamMA: Mamba-based pedestrian trajectory prediction
The second catalog paper predicts pedestrian trajectories with a Mamba state-space model that consumes LiDAR occupancy grid maps and pedestrian awareness states; Evolve rates reproducibility medium, flagging large reported SOTA gains built on narrow baselines and assumed awareness labels.
2026 output surge
The counts-by-year record shows 7 works already in 2026 — the highest annual count in the profile — after 4 in 2025 and 1 in 2024, with 12 works over the most recent three-year window.

Timeline

2014Earliest year in the OpenAlex counts-by-year record, with 1 work; annual output stayed in the low single digits through 2016 (2 in 2015, 1 in 2016).
2017A first peak of 5 works, the highest annual count in the record before 2026, followed by 3 works in 2018 and 4 in 2019.
2021Output settles at 2 works in each of 2021 and 2022, the quietest stretch since 2016, before recovering to 3 in 2023.
2025Output recovers to 4 works, bringing the most recent three-year window to 12 works in total.
2026Two arXiv preprints enter the Evolve catalog — the decoupled early-exits flow-matching VLA paper and MamMA for pedestrian trajectory prediction — and the year already records 7 works, the highest in the profile.

Papers in the Evolve catalog

Frequent co-authors

Riccardo Andrea Izzo×1 Rimvydas Rubavicius×1 Subramanian Ramamoorthy×1 Matteo Matteucci×1 Alessandro Suglia×1 Juncen Long×1 Xiaofeng Jin×1 Simone Mentasti×1 Matteo Matteucci×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 aggregates are computed from OpenAlex only; per-paper citation counts and venue-level detail were not available for this profile.
The output and impact dimensions are marked missing: the current record could not support a reliable score, despite an h-index of 10 and an i10-index of 11 in the raw snapshot.
Evolve's qualitative verdicts and reproducibility ratings cover only the 2 cataloged 2026 arXiv preprints; other works in the counts-by-year record were not reviewed.
Identity is resolved from the provided ORCID and the OpenAlex record; name collisions with other researchers named Gianluca Bardaro cannot be fully excluded.