← Mojtaba Jafarian Abyaneh

Projects

Deep learning for transportation, urban wind simulation, and open-source engineering tools.

Selected Research

Predicting where people and cars go next, from raw LiDARPublished

Where is that pedestrian about to step? A Transformer answers it from raw LiDAR.

  • 12,390+ real trajectories — vehicles, pedestrians, cyclists — from one Ouster OS1 LiDAR at a West Palm Beach intersection.
  • 15.24% lower displacement error than an LSTM baseline, and it converges faster.
  • Full pipeline: point cloud → object detection → curation → geo-referenced prediction.
Aerial map of Clematis Street, West Palm Beach, with a pedestrian's observed path, the ground-truth continuation, and the Transformer and LSTM predicted paths drawn as coloured tracks.
Observed path, ground truth, and Transformer vs. LSTM predictions for one pedestrian outside the Mandel Public Library. Figure from the paper, published open access in IEEE OJ-ITS (CC BY).
Paper · Code

Forecasting bike-share demand a year aheadPreprint

Cyclist risk concentrates where riding concentrates — but the numbers usually arrive after the crash.

  • Predicts station-to-station demand a year ahead with a spatial graph convolutional network.
  • Public data only — trip records, census, OpenStreetMap — so an agency with no data budget can run it.
  • Predicted flow approximates exposure: it points to the corridors and hours where protection pays off.
Map of Jersey City and Hoboken with curved lines between bike-share stations, thickness and colour showing the strength of morning-peak origin–destination flows.
Morning-peak origin–destination flows across the Jersey City Citi Bike network, with stations coloured by land-use zone.
SSRN preprint

Code

Interactive 2D Beam Solver

Drag supports and loads onto a beam. Shear, moment and deflection update live.

A teaching tool for structural analysis — indeterminate beams, 7 support types, 14 worked examples.
One HTML file · no dependencies · works offline
Try it · GitHub

FluidX3D Urban Wind Simulation Guide

GPU lattice-Boltzmann wind flow over real city geometry, start to finish.

Written so the next person can reproduce the run — setup, physics defines, ParaView.
Workflow guide + two reference docs · C++/OpenCL · ParaView
GitHub