Routing, search and scoring, worked out in papers and shipped in products. What we publish, what we talk about, and what you can try.
An invited research presentation on real-time multimodal pathfinding: how unlimited-transfer routing copes with delays, and what it takes to run it inside a live journey planner. Thanks to Hugues Blache for proposing the talk, Maria Lee for organising it, and Meead Saberi for the questions.
For new problems we use new solutions, but some problems require looking back. Over the last decade, BFS-based algorithms such as CSA, RAPTOR and Trip-Based Routing replaced Dijkstra in transit routing, mainly because transfers were short walks. With bicycles and scooters in the mix, that assumption no longer holds. This talk revisits Dijkstra for multimodal transit routing. Joint work with Denys Katkalo and Toby Walsh.
Many transit apps still do not offer full multi-mobility: no long walks, e-scooters or taxis mid-trip, so a journey that exists in the timetable cannot be found in a standard request. This talk shows how to adapt the UnLimited TRAnsfers (ULTRA) framework to delays for RAPTOR and CSA, which makes it practical for real-time use. Next, together with the Karlsruhe Institute of Technology, we plan to extend it to limited transfer radii and re-routing, and to bring it into OpenTripPlanner. Joint work with Denys Katkalo and Toby Walsh.
WCTR happens once every three years and spans freight, logistics, maritime, infrastructure design and transport economics, which puts routing research next to every other angle on how people move. The method presented here already runs in production in OpenTripPlanner, the most widely used open-source transit routing engine. Joint work with Abdallah Abuaisha and Toby Walsh.
Research fields often evolve in one direction: newer replaces older, and we rarely look back. Public transit routing followed this path. In this paper, we take a step back to classical TD-Dijkstra. By applying techniques from computational geometry, we adapt it to handle modern challenges: unlimited transfer distances and personalised travel speeds.
Cars and public transport seem like similar routing problems, but the algorithms that work in each domain are surprisingly different. Contraction Hierarchies, the gold standard for car routing, perform poorly on public transit networks, and vice versa. We tested this gap and showed how computational geometry techniques can help bridge it.
P. Khlaisamniang, K. Kerdthaisong, S. Vorathammathorn, N. Yongsatianchot, H. Phimsiri, T. Thitseesaeng, K. Veerakanjana, K. Kachai, P. Ittichaiwong, T. Saengja
Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2025
A nutritionist-inspired, two-step multimodal-LLM pipeline for food image nutrition analysis.
D. Katkalo, A. Rohovyi, T. Walsh
The IEEE International Conference on Intelligent Transportation Systems (ITSC), 2026
Classical time-dependent Dijkstra, adapted to handle stop buffers and transfers of unlimited distance.
P. Khlaisamniang et al.
The 23rd Workshop on Biomedical Natural Language Processing (BioNLP), ACL 2024
A GPU-poor recipe for radiology report generation with multimodal LLMs.
D. Katkalo, A. Rohovyi, T. Walsh
The Symposium on Combinatorial Search (SoCS), 2026
Multimodal journey planning that stays fast and compact even under real-time delays.
P. Khlaisamniang, S. Phoomvuthisarn
The 35th International Conference on Scientific and Statistical Database Management (SSDBM), Article 8, 2023
A multi-view spatio-temporal graph neural network for traffic forecasting.
A. Rohovyi, A. Abuaisha, T. Walsh
The 16th World Conference on Transport Research (WCTR), 2026
Sort transfers by length and stop scanning early: up to 68% faster transit search, now in OpenTripPlanner.
A. Rohovyi, P.J. Stuckey, T. Walsh
2025 IEEE 37th International Conference on Tools with Artificial Intelligence (ICTAI), 2025
Personalised travel speeds and unlimited-distance transfers, via computational geometry on TD-Dijkstra.
D. Katkalo
2025 IEEE 8th International Conference on Methods and Systems of Navigation and Motion Control (MSNMC), 2025
Band selection and cross-attention for UAV-based hyperspectral image classification.
A. Rohovyi, P.J. Stuckey, T. Walsh
arXiv preprint arXiv:2410.15715, 2024
A graph construction that makes classical point-to-point search competitive on timetable networks.
Yu.V. Maniak, A.V. Rohovyi
International Scientific Journal, 2016
A.V. Rohovyi
International Scientific Journal, 2016
