Robot learning researcher at Midea Group working on world action models for long-horizon, OOD manipulation
3
papers here
3
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
0
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—
OpenAlex lists only 4 works (all 2026), with h-index and i10-index both 0 — too thin to score. This axis would measure publication volume and productivity: total works, annual distribution, and venue spread.
Impact—
No citation evidence supports scoring yet: h-index 0, i10-index 0, and no citation series beyond the 2026 works. This axis would measure scholarly influence — citations, citing works, and their trajectory over time.
Focus0/100
The computed focus score is 0 in this snapshot. The raw field distribution splits evenly (0.5/0.5) between robot manipulation/embodied AI and machine learning for robotics — two tightly adjacent subfields — and OpenAlex topics cluster narrowly (Reinforcement Learning in Robotics 4, Robot Manipulation and Learning 4, two topics at 2 each). With so few works, the low score reflects a very small, young record rather than demonstrated topic-switching, and should be re-read as the record grows.
Collaboration60/100
The collaboration dimension scores 51. OpenAlex records 12 distinct coauthors, and the repeated-collaborator list — Zitai Huang, Taiyi Su, Jian Zhu, Jianjun Zhang, Chong Ma, Weiyi Lu, Yi Xu, Hanli Wang — points to a stable core team behind both 2026 preprints. Authorship roles vary: first author on CSWAM and a middle position on MaskHarness-WAM, neither corresponding, consistent with contributing within a group-led research line rather than independently directing it.
Activity47/100
Activity scores 47. The visible record is concentrated in a single year: countsByYear shows 4 works in 2026 and nothing earlier, and recent3yWorks is 4, matching the two cataloged arXiv preprints plus additional OpenAlex-indexed items. This suggests a recently active or newly indexed profile; with only one year of signal, the score should be read as provisional rather than a verdict on sustained research pace.
Reproducibility32/100
Reproducibility scores 38 (reproMean over 2 reviewed papers). CSWAM is rated medium readiness: it reports +35/42.5 percentage-point gains but no error bars and lacks key robustness baselines. MaskHarness-WAM is rated low: strong real-robot results, yet component ablations are absent. Both gaps are common and fixable; releasing ablations, variance reports, and baseline configurations would materially raise this axis.
Research summary
Tianbin Liu is a robotics researcher at Midea Group (China); this profile is name-resolved from public OpenAlex bibliographic data, with no ORCID linked. The record is early-stage: OpenAlex lists 4 works in 2026, an h-index and i10-index of 0, and 12 distinct coauthors, while Evolve has reviewed two 2026 arXiv preprints. The research program sits in robot manipulation and robot learning — the field distribution splits evenly (0.5/0.5) between robot manipulation/embodied AI and machine learning for robotics — with OpenAlex topics led by Reinforcement Learning in Robotics (4) and Robot Manipulation and Learning (4), plus smaller clusters in AI-based problem solving and planning and multimodal machine learning (2 each). The first-authored CSWAM adds a V-JEPA semantic expert to the FastWAM world action model for out-of-distribution manipulation, reporting gains of +35/42.5 percentage points, though without error bars or key robustness baselines. MaskHarness-WAM, a middle-author work, re-grounds target masks per subtask through a VLM-gated harness for identical-object sequential manipulation, with strong real-robot gains but missing component ablations. Evolve rates reproducibility readiness medium and low respectively (dimension 38), collaboration 51, and activity 47; citation impact is not yet scorable given the nascent record.
Collaboration pattern
Tianbin Liu works within a stable core team: 12 distinct coauthors are recorded, with Zitai Huang, Taiyi Su, Jian Zhu, Jianjun Zhang, Chong Ma, Weiyi Lu, Yi Xu, and Hanli Wang appearing repeatedly. Authorship spans first position on CSWAM and a middle position on MaskHarness-WAM, neither corresponding, suggesting contribution inside a group-led research line at Midea Group rather than independent PI-style leadership.
Research keywords
Highlights
CSWAM: a V-JEPA semantic expert for out-of-distribution world action models
MaskHarness-WAM: instance-grounded harnessing for long-horizon manipulation
Timeline
- 2026OpenAlex records the earliest visible activity: 4 works indexed for the year, spanning robot manipulation/embodied AI and machine learning for robotics, with 12 distinct coauthors.
- 2026First-authored CSWAM preprint appears on arXiv, adding a V-JEPA semantic expert to FastWAM for out-of-distribution generalization in world action models.
- 2026Co-authors MaskHarness-WAM, extending VLM-gated instance-grounded harnessing to long-horizon sequential manipulation of identical objects.
- 2026Evolve reviews both preprints, assigning reproducibility readiness of medium (CSWAM) and low (MaskHarness-WAM), with an overall reproducibility dimension of 38.
Papers in the Evolve catalog
- first author
JEPA-WAM: Connecting Generated Visual Instructions to World Action Models through JEPA Latent Representations
MaskHarness-WAM: Instance-Grounded Harnessing for Long-Horizon Robot Manipulation
- first author
CSWAM: Better Causal Semantic Representations for Out-of-Distribution Generalization in World Action Models
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 OpenAlex only; no ORCID is linked, so this profile is name-resolved from public bibliographic data and may merge or split records for a relatively common Chinese name.
- Only two papers, both 2026 arXiv preprints, are in the Evolve catalog; preprints are not peer-reviewed, and the OpenAlex record may include works not reviewed here.
- Citation aggregates (h-index 0, i10-index 0) reflect a nascent record; the output and impact dimensions are unscored (dimMissing) rather than zero-rated on merit.
- Radar scores are Evolve-computed 0-100 values from the current snapshot; focus and activity in particular rest on a single-year sample and should be re-read as the record matures.