CONTEXT-DEPENDENT TRAJECTORY FORECASTING FOR AUTONOMOUS TRANSPORT SYSTEMS USING HYBRID GRAPH AND TRANSFORMER NEURAL NETWORKS
https://doi.org/10.53360/2788-7995-2026-3(23)-21
Abstract
Trajectory forecasting is a critical component of autonomous transport systems because it supports conflict assessment and safe motion planning in dynamic road environments. This study proposes a hybrid forecasting architecture that combines adaptive graph-based modeling of local agent interactions with a transformer module that encodes the global scene context. The model was evaluated on the Argoverse 2 Motion Forecasting and Waymo Open Motion Dataset benchmarks using ADE, FDE, Miss Rate, minADE, and minFDE, together with analyses of robustness, interaction density, and inference time. On Argoverse 2, the proposed model reduced ADE from 1.21 to 1.17, FDE from 2.81 to 2.69, and Miss Rate from 0.25 to 0.23 compared with Motion Transformer. On Waymo, the improvement was smaller but consistent: ADE decreased from 1.58 to 1.53 and FDE from 3.82 to 3.69. The largest gain was observed in high-density interaction scenes, where ADE decreased from 1.52 to 1.45. These results indicate that adaptive interaction weighting improves forecasting robustness in complex multi-agent road scenarios, although it increases computational cost.
About the Authors
I. BainazarovaKazakhstan
Indira Bainazarova – PhD student, Department of Automation and Control
Tauke Khan Avenue, 5, Shymkent, Kazakhstan
Zh. Umarova
Kazakhstan
Zhanat Umarova – PhD, Associate Professor, Department of Information Systems and Modeling
Tauke Khan Avenue, 5, Shymkent, Kazakhstan
Yu. Jusoh
Malaysia
Yusmadi Jusoh – Associate Professor, Faculty of Computer Science and Information Technology
Serdang, Selangor, Malaysia
Zh. Iztayev
Kazakhstan
Zhalgasbek Iztayev – Candidate of pedagogucal sceinces, Associate Professor, Department of Information Systems and Modeling
Tauke Khan Avenue, 5, Shymkent, Kazakhstan
S. Botayeva
Kazakhstan
Saule Botayeva – Candidate of technical sceinces, Department of IT and Digital Technologies
Konayev Avenue, 16, Shymkent, Kazakhstan
References
1. Social LSTM: Human trajectory prediction in crowded spaces / A. Alahi et al // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). – Las Vegas, USA, 2016. – P. 961-971. https://doi.org/10.1109/CVPR.2016.110.
2. Social-STGCNN: A social spatio-temporal graph convolutional neural network for human trajectory prediction / A. Mohamed et al // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). – Seattle, USA, 2020. – P. 14424-14432. https://doi.org/10.1109/CVPR42600.2020.01444.
3. Learning lane graph representations for motion forecasting / M. Liang et al // Proceedings of the European Conference on Computer Vision (ECCV). – Glasgow, United Kingdom, 2020. – P. 541-556. https://doi.org/10.1007/978-3-030-58536-5_32.
4. AgentFormer: Agent-aware transformers for socio-temporal multi-agent forecasting / Y. Yuan et al // Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). – Montreal, Canada, 2021. – P. 9813-9823. https://doi.org/10.1109/ICCV48922.2021.00967.
5. Motion Transformer with global intention localization and local movement refinement / S. Shi et al // Advances in Neural Information Processing Systems. – 2022. – Vol. 35. – New York: Curran Associates. – P. 6531-6543. https://doi.org/10.48550/arXiv.2209.13508.
6. nuScenes: A multimodal dataset for autonomous driving / H. Caesar et al // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). – Seattle, USA, 2020. – P. 11618-11628. https://doi.org/10.1109/CVPR42600.2020.01164.
7. Zhan W. et al. INTERACTION dataset: An INTERnational, adversarial and cooperative motion dataset in interactive driving scenarios with semantic maps [The electron. resource]. – 2019. https://arxiv.org/abs/1910.03088 (date of request: 17.05.2025). https://doi.org/10.48550/arXiv.1910.03088.
8. Argoverse 2: Next generation datasets for self-driving perception and forecasting / B. Wilson et al // Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks. – 2023. https://arxiv.org/abs/2301.00493 (date of request: 17.05.2025). https://doi.org/10.48550/arXiv.2301.00493.
9. Large scale interactive motion forecasting for autonomous driving: The Waymo Open Motion Dataset / S. Ettinger et al // Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). – Montreal, Canada, 2021. – P. 9710-9719. https://doi.org/10.1109/ICCV48922.2021.00957.
10. VectorNet: Encoding HD maps and agent dynamics from vectorized representation / J. Gao et al // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). – Seattle, USA, 2020. – P. 11525-11533. https://doi.org/10.1109/CVPR42600.2020.01154.
Review
For citations:
Bainazarova I., Umarova Zh., Jusoh Yu., Iztayev Zh., Botayeva S. CONTEXT-DEPENDENT TRAJECTORY FORECASTING FOR AUTONOMOUS TRANSPORT SYSTEMS USING HYBRID GRAPH AND TRANSFORMER NEURAL NETWORKS. Bulletin of Shakarim University. Technical Sciences. 2026;(3(23)):205-216. https://doi.org/10.53360/2788-7995-2026-3(23)-21
JATS XML















