Researcher at University of British Columbia working across robot simulation and evaluation of self-evolving LLM agents
3
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
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—
No publication-volume data is available for this profile yet. The output axis would measure total papers, publication venues and yearly output counts; until the bibliometric snapshot is expanded, this axis is left unscored rather than estimated.
Impact—
Impact data such as citation counts, h-index or venue influence is not available in the provided snapshot. This axis would measure citations and downstream uptake of the reviewed work; it is reported as missing rather than guessed.
Focus0/100
The focus score of 0 reflects an even split in the raw field distribution: 50% robotics (simulation and manipulation) and 50% artificial intelligence (LLM agent evaluation). The three catalogued 2026 papers alternate between the two lines — EvoPathBench on agent evaluation, DiagGen on deformable simulation and SkillZip Pro on agent skills — indicating a deliberately two-track portfolio rather than a single narrow specialty.
Collaboration63/100
Collaboration scores 63, based on 23 distinct coauthors in the bibliometric record, with twelve recurring partners including Xiaofan Bai, Hongqiang Lin, Xuan Jin and Xipeng Cao. All three reviewed 2026 papers list Chao Liu in a middle author position without corresponding role, suggesting a steady contributor embedded in larger teams rather than a solo or consistently leading investigator.
Activity—
Activity data such as yearly publication cadence or recent output velocity is not present in this snapshot. The axis would measure sustained research throughput over time; three reviewed papers all dated 2026 is not, by itself, enough to score it.
Reproducibility60/100
Reproducibility averages 60 across the 2 papers with readiness assessments. Dispersion is wide: the EvoPathBench process-level evaluation paper is rated high — it evaluates at checkpoints and reports negatives honestly — while DiagGen is rated low, being small-n, VLM-judged and without significance tests, and SkillZip Pro is medium, resting on small, private evaluations.
Research summary
Chao Liu is a researcher at the University of British Columbia whose catalogued work spans two equally weighted lines: robotics (simulation and manipulation) and artificial intelligence (LLM agent evaluation) — the field distribution in the bibliometric record is split 50/50 between them. The Evolve catalog holds three 2026 arXiv papers, all with Chao Liu in a middle author position and none as corresponding author. Beyond Endpoint Performance (EvoPathBench) introduces checkpoint-based, process-level evaluation of self-evolving agents and is rated high for reproducibility, with reviewers noting its honesty about negative results alongside single-domain scope and absent error bars. DiagGen builds an agentic pipeline that turns images into simulatable deformable assets via simulation-based, stage-routed repair; it is rated low for reproducibility owing to small samples, VLM-based judgement and no significance testing. SkillZip Pro claims evaluation-free compression of progressively loaded skill bundles, cutting bundle size by 38% without quality loss; it is rated medium because the evidence rests on small, private evaluations. The record lists 23 distinct coauthors with a recurring core of twelve collaborators and a collaboration score of 63. Reproducibility averages 60 over two assessed papers, while focus scores 0, reflecting the perfectly even two-field split. Output, impact and activity metrics are not yet available, so this profile should be read as an early-stage bibliometric snapshot.
Collaboration pattern
The record shows 23 distinct coauthors and a stable recurring core of twelve, including Xiaofan Bai, Hongqiang Lin, Xuan Jin, Xipeng Cao and Yuhong Li. A collaboration score of 63 sits mid-range, and all three reviewed 2026 papers place Chao Liu as a middle author without corresponding role — a pattern consistent with a contributing specialist within larger teams rather than a solo or consistently leading investigator.
Research keywords
Highlights
Checkpoint-based process evaluation for self-evolving agents
Image-to-simulation pipeline for deformable robot assets
Evaluation-free compression of progressively loaded agent skills
Timeline
- 2026Released SkillZip Pro on execution-aware dynamic compression of progressively loaded skills for self-evolving agents (arXiv).
- 2026Released DiagGen on agentic generation of deformable assets with simulation-based diagnostics for robotic simulation (arXiv).
- 2026Released Beyond Endpoint Performance (EvoPathBench), a process-level evaluation framework for self-evolving agents (arXiv).
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.
- Name resolution: 'Chao Liu' is a common name; this profile is resolved from public bibliographic data and may not capture every work by the intended person.
- Bibliometric scores derive from a single public bibliometric snapshot; no citation counts, h-index or venue-level statistics were provided.
- Output, impact and activity axes are unscored because their underlying data is missing from the snapshot.
- All three reviewed papers are 2026 arXiv preprints with middle-author positions and no corresponding role; the reproducibility score covers only 2 papers.