2026-09-05

7 arXiv categories· 96 journals· 2367 candidates — 5 worth your time· 110 without an open abstract

Headline

A transformer-based trajectory prediction framework for varying sensor frame rates and horizons

Today, together

tag shift
no tag ran above its 30-day average
on the same shoulders
7 of the day's papers cite Travel demand and the 3Ds: Density, diversity, and design (Robert Cervero, Kara M. Kockelman 1997); 6 cite Accessibility evaluation of land-use and transport strategies: review and research directions (Karst Geurs, Bert van Wee 2003); 6 cite Travel and the Built Environment (Reid Ewing, Robert Cervero 2010) counted over the 111 of this day's 115 named papers whose reference lists we have — standing on the same older work is not the same as being about the same thing
first appearance
Badr University in Cairo; Capital Health; Laboratoire Traitement et Communication de l’Information; and 1 more none of them on a paper we published in the 85 days before this issue

Two counts, not one counted twice: every row here but “on the same shoulders” counts the papers published above, and that one counts every paper this day named — those and the ones under Also published today, whose abstracts we could not read.

OmniTraj introduces a Transformer-based trajectory prediction framework designed to handle varying sensor frame rates and observation horizons. Another study combines the Analytical Hierarchy Process, Genetic Algorithms, and ArcGIS Pro's Location-Allocation tool to optimize neighborhood service accessibility. Finally, ButterMamba improves traffic flow prediction by combining a Butterworth Spectral Filtering module to filter out high-frequency sensor noise with a parallel Mamba-based Spatial-Temporal State Mixer.

codedatapublished: Transportation Research Part C Emerging Technologies

Yang Gao, Po‐Chien Luan, Kaouther Messaoud, et al.
Transportation Research Part C Emerging Technologies · was arXiv

École Polytechnique Fédérale de Lausanne

Autonomous Vehicle Technology and Safety · Traffic and Road Safety · Gait Recognition and Analysis · self-supervised learning · École Polytechnique Fédérale de Lausanne · Télécom Paris

OmniTraj: Pre-training on heterogeneous data for adaptive and zero-shot human trajectory prediction

The researchers developed OmniTraj, a Transformer-based trajectory prediction framework designed to handle varying sensor frame rates and observation horizons. By explicitly conditioning the model on temporal metadata, it addresses temporal mismatches across datasets and reduces prediction errors by over 70% under joint scene-temporal transfer conditions, while maintaining high accuracy in edge cases with limited observations.

Why it matters — It provides a trajectory prediction model capable of zero-shot generalization across different sensor configurations and unseen environments, removing the need to retrain models when deploying them on vehicles with different camera frame rates or observation windows.

Transportation Research Part C Emerging Technologies · doi · code

published: Computational Urban Science

Mohamed ElZaghal, Mohamed Marzouk
Computational Urban Science

Cairo University

Urban Transport and Accessibility · Urban Design and Spatial Analysis · Place Attachment and Urban Studies · Cairo University · Capital Health · Badr University in Cairo

Optimizing service accessibility using GIS-driven and genetic algorithms for sustainable neighborhoods

This study develops a three-stage framework to optimize neighborhood service accessibility by combining the Analytical Hierarchy Process (AHP), Genetic Algorithms (GAs), and ArcGIS Pro's Location-Allocation tool. After identifying 27 essential urban services from nine sustainable cities and weighting them via expert consultation, the framework was applied to a residential district in Egypt's new administrative capital. The resulting optimized service placements achieved up to a 76% improvement in accessibility and walkability compared to the initial urban design.

Why it matters — It provides urban planners with a systematic, computationally optimized method to locate neighborhood services, shifting from intuitive or standard-based placement to mathematically verified, walkability-focused layouts.

Caveat: While the framework's criteria were derived from nine global sustainable cities, its practical optimization performance was demonstrated on only a single residential district in Egypt.

Computational Urban Science · doi

datapreprint

Limiao Zhang, Yuhui Lu, Jie Gao, et al.
arXiv physics.soc-ph

Traffic Prediction and Management Techniques · Traffic control and management · Human Mobility and Location-Based Analysis

ButterMamba: Butterworth-Enhanced Spatial-Temporal Mamba for Efficient Traffic Flow Prediction

The researchers developed ButterMamba, a traffic flow prediction framework that combines a Butterworth Spectral Filtering module to filter out high-frequency sensor noise with a parallel Mamba-based Spatial-Temporal State Mixer. The model was evaluated on three public datasets, demonstrating linear computational complexity alongside reduced training times and memory consumption compared to existing state-of-the-art models.

Why it matters — It overcomes the quadratic computational complexity of attention-based architectures and the noise sensitivity of standard graph neural networks, enabling real-time, resource-efficient traffic forecasting across large-scale metropolitan road networks.

arXiv · pdf

datapreprint

Etcharla Revanth Rao, Priyanshu Karmakar, Shubhojit Mallick, et al.
arXiv cs.AI

UTP-Bench: Uncertainty-aware Travel Planning Benchmark

The researchers developed UTP-Bench, a benchmark for evaluating how well Large Language Models generate travel itineraries that withstand real-world disruptions. The benchmark incorporates real-world travel data from 504 Indian cities, including multi-modal transport networks, restaurants, and hotels, alongside empirical delay distributions and crowd-density patterns. It introduces three metrics—Buffer Adequacy Score, Crowd-Aware Timing Score, and Transport Delay Absorption Score—to test models like GPT-5, Qwen3, Mistral, and Phi-4 under stochastic conditions.

Why it matters — It shifts the evaluation of LLM-based travel planners from static constraint satisfaction to dynamic robustness, measuring for the first time how automated itineraries handle unpredictable transit delays and crowd fluctuations.

arXiv · pdf

preprint

Kaizhen Tan
arXiv cs.CV

Urban Green Space and Health · Land Use and Ecosystem Services · Remote-Sensing Image Classification · street view imagery · United States

You Cannot Photograph the Same Street Twice: Reliability Limits in Vision-Language Measurement of Urban Change

The study evaluated the longitudinal reliability of vision-language models in measuring urban change using 4,648 consecutive-epoch image pairs from 435 Google Street View standpoints across five US cities. Re-photographing the same unchanged street altered perception scores by an average of 0.80 points, which represents 66.5% of the typical difference between two distinct streets in the same city. While camera geometry in crowdsourced imagery caused models to falsely report physical change in 45% of identical-scene pairs, normalising images to a common virtual camera reduced this error rate to 7.5%.

Why it matters — It establishes that vision-language models are unreliable for detecting urban change at individual sample points due to high sensitivity to acquisition conditions and camera geometry, but demonstrates that aggregating hundreds of paired observations can successfully recover a coherent redevelopment signal.

Caveat: The findings are based on perception scores from five US cities and specific vision-language model architectures, which may not generalize to all geographic contexts or alternative model classes.

arXiv · pdf

Still cited

Geoff Boeing (2017), OSMnx: New methods for acquiring, constructing, analyzing, and visualizing complex street networks 1 of today's items cite it · 29 of 3240 in the archive stand on it

Also published today

These appeared today in journals we track. Their abstracts are not openly available, so we cannot summarise them.