Robot learning and LiDAR perception researcher probing what vision-language and LiDAR-language models truly understand
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
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: this snapshot provides no publication-volume metrics (total outputs, yearly counts, or venue record) beyond the two arXiv papers Evolve has reviewed, so overall research productivity cannot be estimated.
Impact—
Insufficient data: no citation counts, h-index, or reference-impact records are available in this snapshot, so the influence of this researcher's work cannot yet be measured or scored.
Focus0/100
Focus scores 0 on the 0-100 scale, consistent with the raw field distribution: an even 50/50 split between two areas — robotics / robot learning with vision-language-action models, and computer vision for LiDAR scene understanding in autonomous driving. The two catalog papers sit on opposite sides of this divide, so the low score reflects breadth across two multimodal directions rather than concentration in a single niche, not an absence of research activity.
Collaboration58/100
Collaboration scores 58 out of 100, based on 17 distinct coauthors recorded across the reviewed work. The frequent-coauthor list is topped by Kunyu Peng, the only name repeated in the record, alongside Zhiyuan Gao, Yanxiang Zhan, Mohammad Khoshnazar, Runyi Yang, Ruiping Liu, Yufan Chen, Murat Akkoyun, Jeroen Schäfer and Michael Beetz. On both catalog papers Di Wen appears as a middle author rather than corresponding author, consistent with a contributor role inside a larger team.
Activity—
Not yet available: this axis would measure publication cadence and research continuity over time, but the snapshot contains only 2026-dated records, too few data points to gauge an ongoing output rhythm.
Reproducibility78/100
Reproducibility is the strongest measured axis at 78/100, averaged over two papers with reproducibility assessments. The LiDAR-Hallu benchmark study is rated high readiness — a rigorous evaluation protocol on nuScenes that others can rerun — while FOCAL-VLA is rated medium, with solid RoboCasa and real-robot gains but thin statistics. Together they suggest an orientation toward evaluable, benchmark-driven work, though the mean rests on only two data points.
Research summary
Di Wen is an early-stage researcher working at the intersection of robot learning and computer vision, with a catalog profile that splits evenly between vision-language-action (VLA) models for robot manipulation and LiDAR scene understanding for autonomous driving. Two arXiv papers reviewed by Evolve, both from 2026, define this record. The first, FOCAL-VLA, distills VGGT geometry features and Track4World motion representations into the π0.5 policy, restricted to subtask-relevant image regions, and reports solid gains on RoboCasa and real-robot tasks — though with thin statistical support. The second introduces LiDAR-Hallu, a benchmark that exposes answer-bias and relation-blindness in two B4DL LiDAR language models on nuScenes, using a rigorous protocol that tests only one model family. Across both papers Di Wen appears as a middle author rather than corresponding author. Bibliometric data records 17 distinct coauthors, with Kunyu Peng the most frequently recurring, and an even 50/50 split between the two fields. Reproducibility is the strongest measured dimension (78/100 across two papers, with benchmark and distillation readiness rated high and medium respectively), while collaboration scores 58/100. Output, impact and activity metrics are not yet available in this snapshot, so citation-based claims are not made here. The profile is name-resolved from public bibliographic data via ORCID 0009-0000-1693-7912.
Collaboration pattern
Di Wen's recorded network spans 17 distinct coauthors across the two reviewed papers, with Kunyu Peng the only recurring name in the frequent-coauthor list — appearing twice versus single appearances for Zhiyuan Gao, Yanxiang Zhan, Mohammad Khoshnazar, Runyi Yang, Ruiping Liu, Yufan Chen, Murat Akkoyun, Jeroen Schäfer and Michael Beetz. Holding middle-author positions on both papers with no corresponding authorship, the pattern suggests a contributing member of a multi-person robotics and vision team rather than an independent lead.
Research keywords
Highlights
LiDAR-Hallu: a benchmark exposing hallucination in LiDAR language models
FOCAL-VLA: subtask-guided geometry distillation for robot manipulation
Timeline
- 2026Published LiDAR-Hallu, a benchmark probing the spatio-temporal understanding of B4DL LiDAR language models on nuScenes (middle author, arXiv).
- 2026Released FOCAL-VLA, distilling VGGT geometry and Track4World motion into the π0.5 policy for robot manipulation (middle author, arXiv).
- 2026Both reviewed papers appeared in the same year, splitting evenly between vision-language-action robotics and LiDAR scene understanding.
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 and radar dimensions are computed from OpenAlex-sourced bibliometric data; the Evolve catalog holds only two arXiv papers (both 2026) for this researcher.
- No institution or country metadata is available; the profile is name-resolved from public bibliographic records under ORCID 0009-0000-1693-7912, and "Di Wen" is a common Chinese name.
- Output, impact and activity axes lack underlying data (no citation counts, publication totals, or multi-year records), so those dimensions are reported as missing rather than scored.
- Reproducibility (78/100) is averaged over only two assessed papers, and both are arXiv-sourced without peer-review status.