Robot-learning researcher at Tongji University working on vision-language-action models for real-robot manipulation.
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
4
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 underlying data is insufficient. This axis would measure total publication volume, venue spread, and authorship position across the indexed record; the three 2026 preprints alone do not support a reliable score.
Impact—
Not yet scored: citation-based evidence has not accumulated for this early-stage profile. This axis would measure citations, citation trajectory, and citing works; the platform has no computed value yet, so no estimate is offered.
Focus0/100
Scores lowest of the computed dimensions (0/100). The OpenAlex field distribution splits evenly — one third each for robot manipulation/embodied AI, vision-language-action models, and machine learning for robot manipulation — while topics spread across multimodal ML applications (13), reinforcement learning in robotics (8), robot manipulation and learning (8), human pose and action recognition (6), and neural network applications (5). In practice these labels all orbit manipulation learning, but the index records breadth rather than a single dominant field.
Collaboration60/100
Scores 60/100. The profile lists 19 distinct coauthors across 14 works in the last three years, with recurring partners including Zitai Huang, Tianbin Liu, Jian Zhu, Jianjun Zhang, Chong Ma, Yi Xu, Weiyi Lu, and Hanli Wang. All three reviewed papers place Su as a middle-list, non-corresponding author, indicating a contributor role within a stable team rather than an independent lead.
Activity79/100
The strongest dimension (79/100). Indexed works per year run 1 (2020), 2 (2022), 5 (2023), 2 (2024), 3 (2025), and 9 in 2026 — the highest annual count on record — with 14 of 22 indexed works falling in the last three years. Three arXiv preprints already appear in 2026, confirming a clearly accelerating cadence.
Reproducibility32/100
Scores 32/100 (reproducibility mean across 3 reviewed papers). Evolve's verdicts are concrete: JEPA-WAM ships no code, task-agnostic control, or significance tests; MaskHarness-WAM lacks component ablations; CSWAM reports +35/42.5pp gains without error bars and omits key robustness baselines. Two papers rate low and one medium on repro readiness.
Research summary
Taiyi Su is a robotics researcher at Tongji University whose published work centers on robot learning, specifically vision-language-action (VLA) and world action models (WAMs) for real-robot manipulation. The Evolve-reviewed record comprises three 2026 arXiv preprints, each a systems contribution to the fast-moving WAM ecosystem. JEPA-WAM connects text-to-image-generated visual instructions to world action models through V-JEPA latent representations and reports strong real-robot wins over Fast-WAM; MaskHarness-WAM introduces a VLM-gated, instance-grounded harness that re-grounds target masks per subtask for long-horizon manipulation of identical objects; CSWAM adds a V-JEPA-based causal semantic expert to FastWAM, lifting out-of-distribution performance by 35 and 42.5 points. On all three Su appears as a middle-list, non-corresponding author, consistent with a contributing role in a larger team. Bibliometric coverage is early-stage: h-index 4, i10-index 4, 22 indexed works since 2020, 14 within the last three years, and a marked acceleration to 9 works in 2026. OpenAlex topics attribute 13 works to multimodal machine learning applications and 8 each to reinforcement learning in robotics and robot manipulation and learning. Activity is the strongest computed dimension (79); reproducibility is weak (32) because reviews flag missing code, ablations, error bars, and significance tests. Citation-based output and impact scores are not yet computed.
Collaboration pattern
Su publishes with a stable, sizable team: 19 distinct coauthors over the last three years, with recurring names such as Zitai Huang, Tianbin Liu, Jian Zhu, Jianjun Zhang, Chong Ma, Yi Xu, Weiyi Lu, and Hanli Wang. Across all three reviewed 2026 papers Su sits in the middle of the author list without corresponding-author status, consistent with a technical contributor role in a lab-driven WAM research line rather than solo or senior-lead authorship.
Research keywords
Highlights
JEPA-WAM: generated visual instructions drive world action models
Instance-grounded harness for long-horizon robot manipulation
CSWAM lifts out-of-distribution manipulation by 35-42.5 points
Timeline
- 2020First indexed work appears in the OpenAlex record, with 1 work in 2020.
- 2023Most productive pre-2026 year with 5 indexed works.
- 20253 indexed works, maintaining steady output ahead of the 2026 acceleration.
- 2026Most active year: 9 indexed works, including three WAM-related arXiv preprints reviewed by Evolve (JEPA-WAM, MaskHarness-WAM, CSWAM).
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 figures come from an OpenAlex identity record; citation-based output and impact scores are not yet computed.
- All three reviewed papers are 2026 arXiv preprints; peer-reviewed venue status is not reflected in the data.
- Only three papers have been reviewed by Evolve; topic and yearly counts cover a broader indexed record not individually assessed.
- The profile is name-resolved from public bibliographic data linked to ORCID 0000-0002-4357-1095; biographical details beyond affiliation are not covered.