OnSiteVRU: A High-Resolution Trajectory Dataset for High-Density Vulnerable Road Users

Jianqiang Li; Jian Sun; Peng Hang; Zhangcun Yan

REVIEW 1 major objections 1 minor 1 cited by

Reviewed by Pith at T0; open to challenge.

T0 means a machine referee read the full paper against a public rubric. The mark states how deep the mechanical check went, never who wrote it. the ladder, T0–T4 →

Challenge this review Download PDF Read on arXiv ↗

T0 review · grok-4.3

The OnSiteVRU datasets supply 17,429 trajectories of motor vehicles, electric bicycles, and human-powered bicycles at 0.04-second resolution across intersections, road segments, and urban villages.

2026-05-22 22:16 UTC

load-bearing objection A data release paper for a new Shanghai VRU trajectory set captured with aerial plus onboard sensors; claims higher density but shows little validation detail. the 1 major comments →

arxiv 2503.23365 v3 submitted 2025-03-30 cs.CV cs.RO

OnSiteVRU: A High-Resolution Trajectory Dataset for High-Density Vulnerable Road Users

Zhangcun Yan , Jianqiang Li , Peng Hang , Jian Sun This is my paper

classification cs.CV cs.RO

keywords trajectory datasetvulnerable road usersmixed traffichigh-resolution trajectoriesurban driving scenariosaerial and onboard dataautonomous driving testing

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper creates the OnSiteVRU datasets to supply trajectory data that existing collections lack for dense VRU behavior in mixed traffic. It combines aerial-view recordings of natural driving with onboard real-time detection to capture vehicles and bicycles together with signals, obstacles, and maps. The resulting collection reaches higher VRU density and wider scene coverage than prior sets, which the authors state will improve traffic flow models, prediction methods, and virtual testing of autonomous systems.

Core claim

OnSiteVRU integrates aerial-view natural driving data and onboard real-time dynamic detection data to reconstruct interaction events, yielding approximately 17,429 trajectories at 0.04-second precision for motor vehicles, electric bicycles, and human-powered bicycles in intersections, road segments, and urban villages, together with environmental information such as traffic signals, obstacles, and real-time maps.

What carries the argument

The OnSiteVRU dataset, formed by fusing aerial-view natural driving recordings with onboard real-time dynamic detection to reconstruct full interaction events.

Load-bearing premise

The aerial and onboard sources together record every relevant interaction without large measurement error, occlusion gaps, or timing mismatches.

What would settle it

A direct count showing that OnSiteVRU does not exceed the VRU density or scene variety of existing public trajectory datasets in comparable urban settings.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

Traffic flow models can incorporate observed bicycle-vehicle interactions at higher spatial and temporal detail.
Trajectory prediction algorithms gain training examples that include dense VRU groups and signal-controlled crossings.
Virtual testing environments for autonomous vehicles receive more representative mixed-traffic scenarios.
Behavioral studies of cyclists and drivers in urban villages become feasible with synchronized environmental context.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Cities with similar mixed-traffic patterns could adopt the same dual-collection method to build localized datasets.
The data format may support direct comparison of interaction rates before and after infrastructure changes such as bike lanes.
Prediction models trained on this set could be tested for transfer to pedestrian-only or highway settings to measure domain shift.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

A data release paper for a new Shanghai VRU trajectory set captured with aerial plus onboard sensors; claims higher density but shows little validation detail.

read the letter

This paper releases OnSiteVRU, a trajectory dataset for motor vehicles, e-bikes, and bikes collected in Shanghai intersections, road segments, and urban villages. It combines aerial natural driving footage with onboard real-time detection, reports about 17,429 trajectories at 0.04-second intervals, and includes traffic signals, obstacles, and maps. The data sits on Kaggle for public download. That combination of sources and the focus on high-VRU-density mixed traffic in specific Shanghai settings is the concrete new element. Prior datasets exist, but this one targets scenarios that may be underrepresented elsewhere. The paper does a service by making the raw trajectories and context available rather than keeping them private. Researchers working on prediction models or virtual testing for autonomous driving in dense urban areas can download and use it directly. The soft spot is the thin support for the performance claims. The abstract states that the data outperforms traditional sets in VRU density and scene coverage, yet it supplies no error metrics, fusion accuracy checks, synchronization details, or side-by-side tables. Without those, it is hard to know whether occlusion handling or multi-view alignment actually delivers the stated precision across all claimed scenarios. The full text might contain the missing comparisons, but the version reviewed leaves that gap. This work is aimed at groups that need fresh VRU trajectory material for model training or safety validation. A reader focused on autonomous driving datasets would find it worth examining. It deserves peer review because usable public data releases of this scale can still benefit from external checks on collection quality even when the paper itself is largely descriptive.

Referee Report

1 major / 1 minor

Summary. The paper introduces the OnSiteVRU dataset, a multi-source collection of high-resolution (0.04 s) trajectory data for ~17,429 trajectories involving motor vehicles, electric bicycles, and human-powered bicycles across intersections, road segments, and urban villages. Data are obtained by fusing aerial-view natural driving recordings with onboard real-time dynamic detection, augmented by traffic signals, obstacles, and map information; the work claims superior VRU density and scene coverage relative to prior datasets and releases the data publicly via Kaggle.

Significance. A rigorously validated release of this scale and temporal resolution in high-density mixed-traffic settings would supply a useful resource for trajectory prediction, interaction modeling, and virtual testing of autonomous systems, particularly where existing corpora under-represent VRU behaviors.

major comments (1)

[Abstract] Abstract: the stated 0.04 s temporal precision, total trajectory count, and outperformance in VRU density/scene coverage are asserted without accompanying quantitative error metrics, synchronization validation, or explicit comparison tables/methodology against reference datasets; this directly affects the central claim that the released data are demonstrably superior.

minor comments (1)

The manuscript should include at least one table or figure summarizing per-scenario statistics (e.g., trajectory counts, average speeds, occlusion rates) to allow readers to assess coverage claims directly.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment and the recommendation for minor revision. We address the concern regarding the abstract below.

read point-by-point responses

Referee: [Abstract] Abstract: the stated 0.04 s temporal precision, total trajectory count, and outperformance in VRU density/scene coverage are asserted without accompanying quantitative error metrics, synchronization validation, or explicit comparison tables/methodology against reference datasets; this directly affects the central claim that the released data are demonstrably superior.

Authors: We agree that the abstract, as a concise summary, would benefit from clearer linkage to supporting evidence. The manuscript provides quantitative error metrics for the 0.04 s temporal precision, synchronization validation between aerial and onboard sources, the exact trajectory count, and explicit comparison tables plus methodology for VRU density and scene coverage against prior datasets in Sections 3 (data acquisition and fusion) and 4 (validation and benchmarking). To address the referee's point directly, we will revise the abstract to reference these sections and incorporate brief quantitative highlights of the comparisons. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is a dataset release paper whose central claims concern the collection, integration, and coverage statistics of trajectory data from aerial and onboard sources. No equations, fitted parameters, predictions, or derivations are present in the abstract or described content. All claims are descriptive and externally verifiable via the released dataset itself rather than reducing to self-referential inputs or self-citations.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Dataset release depends on unstated assumptions about sensor accuracy and data representativeness rather than new mathematical constructs.

axioms (1)

domain assumption Aerial and onboard sensors can be synchronized and calibrated to produce 0.04 s trajectory precision without significant drift or occlusion errors.
Precision claim in abstract rests on this unverified measurement assumption.

pith-pipeline@v0.9.0 · 5785 in / 1173 out tokens · 54297 ms · 2026-05-22T22:16:28.995376+00:00 · methodology

0 comments

read the original abstract

With the acceleration of urbanization and the growth of transportation demands, the safety of vulnerable road users (VRUs, such as pedestrians and cyclists) in mixed traffic flows has become increasingly prominent, necessitating high-precision and diverse trajectory data to support the development and optimization of autonomous driving systems. However, existing datasets fall short in capturing the diversity and dynamics of VRU behaviors, making it difficult to meet the research demands of complex traffic environments. To address this gap, this study developed the OnSiteVRU datasets, which cover a variety of scenarios, including intersections, road segments, and urban villages. These datasets provide trajectory data for motor vehicles, electric bicycles, and human-powered bicycles, totaling approximately 17,429 trajectories with a precision of 0.04 seconds. The datasets integrate both aerial-view natural driving data and onboard real-time dynamic detection data, along with environmental information such as traffic signals, obstacles, and real-time maps, enabling a comprehensive reconstruction of interaction events. The results demonstrate that VRU\_Data outperforms traditional datasets in terms of VRU density and scene coverage, offering a more comprehensive representation of VRU behavioral characteristics. This provides critical support for traffic flow modeling, trajectory prediction, and autonomous driving virtual testing. The dataset is publicly available for download at: https://www.kaggle.com/datasets/zcyan2/mixed-traffic-trajectory-dataset-in-from-shanghai.

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

A knowledge-augmented dataset of high-risk driving scenarios with LLM annotations for autonomous driving
cs.LG 2026-07 conditional novelty 6.0

K-Risk curates 31,398 high-risk driving events from 20 trajectory datasets with multi-layered semantic and LLM-generated annotations validated via closed-loop simulation.