Early-career Tsinghua researcher advancing training-free planning and loss reweighting for world models in embodied AI
3
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
3
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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 OpenAlex snapshot gives only h-index 3 and i10-index 1, so per-work output for this identity is unresolved. This axis would measure publication count, venue tier, and first/corresponding-author share.
Impact—
Not yet scored: per-paper citation counts are absent from the snapshot (h-index 3 and i10-index 1 are the only signals), so influence cannot be fairly quantified. This axis would measure citations, downstream adoption, and benchmark uptake of the world-model work.
Focus100/100
This axis scores 100. OpenAlex topics are tightly concentrated on prediction: Time Series Analysis and Forecasting (10) and Forecasting Techniques and Applications (9) lead, extended by Traffic Prediction and Management (5), Reinforcement Learning in Robotics (4), and Stock Market Forecasting (4). The three Evolve-reviewed 2026 papers all target world models for embodied and robot learning, matching the single fieldDist entry.
Collaboration59/100
This axis scores 59. The snapshot records 18 distinct co-authors across 14 works in the last three years — a broad but mid-density network rather than one dominant cluster. Recurring partners include Rongze Tang, Jianjie Fang, Zhaolu Wang, Ziyou Wang, and Haisheng Su, with Chen Gao, Yong Li, Wei Wu, and Zhibo Chen also among frequent collaborators, consistent with Tsinghua-based teams.
Activity79/100
This axis scores 79. OpenAlex records 14 works over the last three years, and countsByYear climbs from 5 publications in 2025 to 9 in 2026 — the steepest recent ramp. All three Evolve-reviewed papers appeared as arXiv preprints in 2026, one of them first-authored, indicating an early-career output curve still accelerating.
Reproducibility100/100
This axis scores 100 (reproMean 100), but it rests on 1 formally assessed paper in the snapshot. That paper, first-authored 'Aim Short to Reach Far', is rated high for reproducibility: training-free, planning on top of a released frozen world model across four tasks. The other two cataloged papers carry medium (CAER) and low (IMPACT) readiness, so the perfect mean is a thin-base signal.
Research summary
Xvyuan Liu is an early-career machine-learning researcher at Tsinghua University whose reviewed work centers on world models for embodied and robot learning. The Evolve catalog holds three 2026 arXiv preprints. As first author of 'Aim Short to Reach Far', Liu shows that training-free anchored planning — aiming a frozen LeWM at short recorded subgoals — beats the model's released planner on four tasks, and the paper carries high reproducibility readiness. Two middle-author papers extend world-model training: 'CAER' introduces online causal action-effect reweighting for action-conditioned video generation, and 'IMPACT' uses a cross-attention prior to reweight losses for scalable interaction-aware training. The OpenAlex snapshot records h-index 3, i10-index 1, 14 works in the last three years, and 18 distinct co-authors; its topic profile is concentrated on prediction (time series 10, forecasting techniques 9, traffic prediction 5, robotics RL 4, stock market 4), which diverges from the world-model catalog and suggests the identity is name-resolved from public bibliographic data. Output is accelerating, from 5 works in 2025 to 9 in 2026, and reproducibility scores 100, though based on one formally assessed paper. Output and impact dimensions remain unscored for lack of citation-side data.
Collaboration pattern
Across 14 works in the last three years, Xvyuan Liu has published with 18 distinct co-authors (collaboration score 59). The recurring core includes Rongze Tang, Jianjie Fang, Zhaolu Wang, Ziyou Wang, and Haisheng Su, with Chen Gao, Yong Li, Wei Wu, Zhibo Chen, Zhuohang Li, and Xin Zhang also appearing frequently — a pattern consistent with several Tsinghua-based project teams rather than one fixed group. The first-authored planning paper draws on this same collaborator pool.
Research keywords
Highlights
Training-free anchored planning beats a frozen world model's own planner
Causal action-effect reweighting for world-model training
Cross-attention priors scale interaction-aware world-model training
Timeline
- 2025OpenAlex countsByYear records 5 publications in 2025, the earliest recent activity shown in the snapshot.
- 2026First-authored arXiv preprint 'Aim Short to Reach Far' introduces training-free anchored planning for frozen world models.
- 2026Middle-author preprint 'CAER' proposes causal action-effect reweighting for world-model training.
- 2026Middle-author preprint 'IMPACT' introduces attention-prior reweighting for interaction-aware world-model training.
- 2026countsByYear peaks at 9 publications, the most active year recorded in the snapshot.
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 scores come from an OpenAlex snapshot only (identity: openalex); output and impact dimensions are unscored because per-paper citation data is unavailable.
- The OpenAlex topic profile (time-series, traffic, and stock forecasting) diverges from the Evolve-reviewed world-model papers; this profile is name-resolved from public bibliographic data and may blend identities.
- countsByYear includes a lone 2001 entry (1 work), likely an indexing artifact; it is excluded from the activity narrative.
- All three reviewed papers are 2026 arXiv preprints; peer-review status and final venues are not yet known.