pith. sign in

arxiv: 2605.16442 · v1 · pith:Y3JTDMBSnew · submitted 2026-05-15 · 💻 cs.RO · cs.AI· cs.LG

Hierarchical Two-Stage Framework for Environment-Aware Long-Horizon Vessel Trajectory Prediction

Ganeshaaraj Gnanavel , Tharindu Fernando , Sridha Sridharan , Clinton Fookes This is my paper

Pith reviewed 2026-05-20 19:23 UTC · model grok-4.3

classification 💻 cs.RO cs.AIcs.LG
keywords vessel trajectory predictionlong-horizon forecastinghierarchical two-stage modelenvironment-aware predictionspatio-temporal graph transformermaritime data fusionocean current integration
0
0 comments X

The pith

A hierarchical two-stage framework improves accuracy of long-horizon vessel trajectory predictions by incorporating environmental factors and fusing coarse and fine-grained models.

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

The paper introduces a hierarchical two-stage framework to forecast vessel trajectories over extended periods in dynamic ocean settings. It pairs a long-term branch that learns overall navigational intent with a short-term branch that uses a Spatio-Temporal Graph Transformer on discretized sea cells to handle local movements. Ocean data on currents, wind, and waves is integrated through attention mechanisms, and a smoothing layer ensures smooth outputs. Tested on real Australian vessel data with three hours of input to predict ten hours ahead, it reduces errors compared to existing approaches. Readers interested in maritime safety and logistics would find this relevant because better predictions support collision avoidance and efficient route planning.

Core claim

The authors develop a hierarchical two-stage framework that combines a coarse long-term predictor with a grid-aware short-term predictor through a hierarchical fusion mechanism. The short-term component employs a Spatio-Temporal Graph Transformer on maritime cells, while an environmental module uses cross-modal attention to adapt to sea conditions such as currents, wind vectors, and wave height. A learnable Savitzky-Golay smoothing layer is added to improve temporal coherence, resulting in superior performance on long prediction horizons.

What carries the argument

The hierarchical fusion mechanism that integrates the coarse navigational intent from the long-term branch with the localized dynamics from the Spatio-Temporal Graph Transformer in the short-term branch, adapted via environmental feature modulation.

If this is right

  • Improved collision avoidance capabilities for maritime traffic management systems using more reliable 10-hour ahead forecasts.
  • Enhanced route planning that dynamically responds to changing oceanographic conditions like currents and waves.
  • More effective traffic management through better anticipation of vessel positions over long time spans.
  • The ablation studies confirm that the environmental integration and fusion steps each add measurable value to the prediction quality.

Where Pith is reading between the lines

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

  • This framework could potentially be extended to predict trajectories for other types of vehicles operating in dynamic environments, such as aircraft or autonomous drones.
  • Real-time applications for autonomous shipping might benefit from incorporating this method to make informed navigation decisions based on forecasted paths.
  • Future work could test the model with higher-resolution environmental data or different geographic regions to assess generalizability.

Load-bearing premise

The oceanographic parameters are precisely aligned in both time and space with the vessel position data, allowing the fusion mechanism to combine intent and dynamics without errors.

What would settle it

Training and evaluating the model on a dataset where the timing of environmental data is intentionally shifted by several hours, and checking if the performance advantage over non-environmental baselines disappears.

Figures

Figures reproduced from arXiv: 2605.16442 by Clinton Fookes, Ganeshaaraj Gnanavel, Sridha Sridharan, Tharindu Fernando.

Figure 1
Figure 1. Figure 1: This figure presents a hierarchical two-stage framework designed for long-horizon, environment-aware vessel trajectory prediction. The architecture [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the grid-aware short-term prediction module. The spatial domain is partitioned into 120 grids and vessel trajectories are grouped by [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Map showing the areas of interest We utilized weather and environmental data from the Copernicus Marine Service2 to obtain current environmental conditions over the identical spatial and temporal parameters. We utilized (i) sea-surface current velocity, (ii) sea-surface wind velocity, and (iii) significant wave height. Environmental factors were temporally aligned and spatially matched with AIS observation… view at source ↗
Figure 4
Figure 4. Figure 4: Three representative examples. In each row, the left plot shows the output prior to smoothing, and the right plot shows the corresponding output after [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Three representative examples. In each plot, the blue line shows the output from the original framework, and the red line shows the output after [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
read the original abstract

Long-horizon vessel trajectory forecasting under real ocean conditions is critical for collision avoidance, traffic management, and route planning. However, achieving accurate predictions is challenging due to long-range temporal dependencies and dynamic environmental factors such as currents, wind, and waves. To address these issues, we propose a hierarchical two-stage framework that combines a coarse long-term predictor with a grid-aware short-term predictor through a hierarchical fusion mechanism. The short-term branch leverages a Spatio-Temporal Graph Transformer on discretized maritime cells to capture localized dynamics, while the long-term branch encodes overarching navigational intent. An integrated environmental module incorporates oceanographic parameters, including surface currents, wind vectors, and significant wave height, using cross-modal attention and feature-wise modulation for adaptive response to varying sea conditions. Additionally, a learnable Savitzky-Golay smoothing layer enhances temporal coherence in fused trajectories. We evaluate our approach on Australian Craft Tracking System (CTS) data from the North West region, aligned with Copernicus Marine Service products, using a 3-hour input and a 10-hour prediction horizon. Experimental results show that our framework outperforms the state-of-the-art by 25% in Average Displacement Error (ADE) and 17% in Final Displacement Error (FDE). Ablation studies further validate the contribution of each component.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a hierarchical two-stage framework for long-horizon vessel trajectory prediction that integrates a coarse long-term predictor encoding navigational intent with a grid-aware short-term predictor based on a Spatio-Temporal Graph Transformer operating on discretized maritime cells. Environmental factors (surface currents, wind vectors, significant wave height) from Copernicus Marine Service are incorporated via cross-modal attention and feature-wise modulation, with a learnable Savitzky-Golay smoothing layer for temporal coherence. The approach is evaluated on Australian CTS data from the North West region (3-hour input, 10-hour prediction horizon) and claims to outperform state-of-the-art methods by 25% in ADE and 17% in FDE, with ablation studies validating individual components.

Significance. If the reported gains prove robust, the work would advance environment-aware trajectory forecasting for maritime robotics applications such as collision avoidance and route planning. The hierarchical fusion of intent and local dynamics, combined with explicit oceanographic conditioning, addresses a practically relevant gap; the inclusion of ablation studies is a positive step toward isolating component contributions.

major comments (2)
  1. [Experimental results / evaluation section] The central performance claim (25% ADE / 17% FDE improvement) is load-bearing yet presented without error bars, statistical significance tests, baseline implementation details, or dataset split information. This omission prevents verification that the gains are not artifacts of a particular split or random seed and directly undermines confidence in the ablation-validated component contributions.
  2. [Methods / data preparation and environmental module] The environmental module relies on precise spatio-temporal alignment between Copernicus Marine Service parameters and Australian CTS vessel tracks within each 3-hour input window. No explicit description or validation of this alignment process (e.g., interpolation method, temporal tolerance, or handling of missing data) is provided; misalignment would propagate through the cross-modal attention and feature-wise modulation into the 10-hour predictions, potentially inflating results.
minor comments (1)
  1. [Abstract] The abstract states the prediction horizon and input length but does not indicate how many baselines were compared or whether the reported percentages are relative to the best baseline or an average; adding this would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important aspects for improving reproducibility and methodological transparency. We address each major comment below and will revise the manuscript to incorporate the suggested enhancements.

read point-by-point responses

  1. Referee: [Experimental results / evaluation section] The central performance claim (25% ADE / 17% FDE improvement) is load-bearing yet presented without error bars, statistical significance tests, baseline implementation details, or dataset split information. This omission prevents verification that the gains are not artifacts of a particular split or random seed and directly undermines confidence in the ablation-validated component contributions.

    Authors: We agree that the current presentation of results would benefit from additional statistical rigor and implementation details to allow independent verification. In the revised manuscript, we will update the experimental section to report performance as mean and standard deviation over five independent runs with different random seeds, include paired statistical significance tests (e.g., t-tests with p-values) against baselines, specify the dataset split procedure (chronological 70/15/15 train/validation/test split to avoid temporal leakage), and provide full implementation details for all baselines including hyperparameter choices and any public code references. These changes will directly support the robustness of the reported 25% ADE and 17% FDE gains as well as the ablation studies. revision: yes

  2. Referee: [Methods / data preparation and environmental module] The environmental module relies on precise spatio-temporal alignment between Copernicus Marine Service parameters and Australian CTS vessel tracks within each 3-hour input window. No explicit description or validation of this alignment process (e.g., interpolation method, temporal tolerance, or handling of missing data) is provided; misalignment would propagate through the cross-modal attention and feature-wise modulation into the 10-hour predictions, potentially inflating results.

    Authors: We acknowledge that an explicit account of the alignment procedure is required to substantiate the environmental conditioning. We will add a new subsection in the Data Preparation section describing the process in detail: Copernicus Marine Service fields (hourly temporal resolution) are aligned to vessel track timestamps via bilinear spatial interpolation combined with linear temporal interpolation; a maximum temporal tolerance of 30 minutes is enforced for valid matches, with missing or out-of-tolerance values imputed by forward-fill from the nearest valid observation within the 3-hour window or excluded if no valid data exists. We will also report a validation metric, such as the mean alignment error computed over a random sample of tracks, to confirm the procedure does not introduce systematic bias. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical ML framework with external validation

full rationale

The paper presents a hierarchical two-stage neural architecture (coarse long-term predictor + grid-aware short-term Spatio-Temporal Graph Transformer + cross-modal environmental fusion + learnable Savitzky-Golay layer) trained end-to-end on external maritime tracking and oceanographic datasets. Performance numbers (25% ADE / 17% FDE improvement) are reported from held-out evaluation on Australian CTS data aligned with Copernicus products; these are not derived quantities but direct experimental outcomes. No equations, uniqueness theorems, or self-citations are invoked to force the central result. The framework is self-contained against external benchmarks and does not reduce any claimed prediction to its own fitted inputs or prior author work by construction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the effectiveness of the hierarchical fusion and environmental module; the abstract introduces several learnable components whose parameters are not enumerated but must be optimized during training.

free parameters (1)
  • learnable Savitzky-Golay smoothing parameters
    A learnable smoothing layer is introduced to enhance temporal coherence; its coefficients are fitted during training.
axioms (1)
  • domain assumption Oceanographic fields from Copernicus are temporally and spatially aligned with vessel positions
    The environmental module assumes accurate alignment between external marine data products and tracking records.

pith-pipeline@v0.9.0 · 5779 in / 1333 out tokens · 43915 ms · 2026-05-20T19:23:04.139186+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Determining and visualizing safe motion parameters of a ship navigating in severe weather conditions,

    R. Szlapczynski and P. Krata, “Determining and visualizing safe motion parameters of a ship navigating in severe weather conditions,”Ocean Engineering, vol. 158, pp. 263–274, 2018

    work page 2018

  2. [2]

    Vessel trajectory prediction in maritime transportation: Current approaches and beyond,

    X. Zhang, X. Fu, Z. Xiao, H. Xu, and Z. Qin, “Vessel trajectory prediction in maritime transportation: Current approaches and beyond,” IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 11, pp. 19 980–19 998, 2022

    work page 2022

  3. [3]

    Stmgf-net: a spa- tiotemporal multi-graph fusion network for vessel trajectory forecasting in intelligent maritime navigation,

    J. Jiang, Y . Zuo, Y . Xiao, W. Zhang, and T. Li, “Stmgf-net: a spa- tiotemporal multi-graph fusion network for vessel trajectory forecasting in intelligent maritime navigation,”IEEE Transactions on Intelligent Transportation Systems, 2024

    work page 2024

  4. [4]

    Bidirectional data-driven trajectory prediction for intelligent maritime traffic,

    Y . Xiao, X. Li, W. Yao, J. Chen, and Y . Hu, “Bidirectional data-driven trajectory prediction for intelligent maritime traffic,”IEEE Transactions on Intelligent Transportation Systems, vol. 24, no. 2, pp. 1773–1785, 2022

    work page 2022

  5. [5]

    Fundamentals of recurrent neural network (rnn) and long short-term memory (lstm) network,

    A. Sherstinsky, “Fundamentals of recurrent neural network (rnn) and long short-term memory (lstm) network,”Physica D: Nonlinear Phe- nomena, vol. 404, p. 132306, 2020

    work page 2020

  6. [6]

    Long short-term memory,

    A. Graves, “Long short-term memory,”Supervised sequence labelling with recurrent neural networks, pp. 37–45, 2012

    work page 2012

  7. [7]

    Informer-based model for long-term ship trajectory prediction

    C. Xiong, H. Shi, J. Li, X. Wu, and R. Gao, “Informer-based model for long-term ship trajectory prediction.”Journal of Marine Science & Engineering, vol. 12, no. 8, 2024

    work page 2024

  8. [8]

    Mstformer: Motion inspired spatial-temporal transformer with dynamic-aware attention for long-term vessel trajectory prediction,

    H. Qiang, Z. Guo, S. Xie, and X. Peng, “Mstformer: Motion inspired spatial-temporal transformer with dynamic-aware attention for long-term vessel trajectory prediction,”arXiv preprint arXiv:2303.11540, 2023

    work page arXiv 2023

  9. [9]

    Ea-vtp: Environment-aware long-term vessel trajectory prediction,

    Z. Huang, Z. Wang, H. Chen, Z. Zhang, J. Wang, Z. Yuan, Y . Jin, and X. Wu, “Ea-vtp: Environment-aware long-term vessel trajectory prediction,” in2022 International joint conference on neural networks (IJCNN). IEEE, 2022, pp. 1–7

    work page 2022

  10. [10]

    An adaptive multimodal data vessel trajectory prediction model based on a satellite automatic identification system and environmental data,

    Y . Xiao, Y . Hu, J. Liu, Y . Xiao, and Q. Liu, “An adaptive multimodal data vessel trajectory prediction model based on a satellite automatic identification system and environmental data,”Journal of Marine Sci- ence and Engineering, vol. 12, no. 3, p. 513, 2024

    work page 2024

  11. [11]

    Long-term trajectory prediction for oil tankers via grid-based clustering,

    X. Xu, C. Liu, J. Li, Y . Miao, and L. Zhao, “Long-term trajectory prediction for oil tankers via grid-based clustering,”Journal of Marine Science and Engineering, vol. 11, no. 6, p. 1211, 2023

    work page 2023

  12. [12]

    Big data driven vessel trajectory and navigating state predic- tion with adaptive learning, motion modeling and particle filtering techniques,

    Z. Xiao, X. Fu, L. Zhang, W. Zhang, R. W. Liu, Z. Liu, and R. S. M. Goh, “Big data driven vessel trajectory and navigating state predic- tion with adaptive learning, motion modeling and particle filtering techniques,”IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 4, pp. 3696–3709, 2020

    work page 2020

  13. [13]

    Framework for ship trajectory forecasting based on linear stationary models using automatic identification system,

    S. Srivastava, L. Kumar, R. Jeyanthi, K. Deepa, and V . Aggrawal, “Framework for ship trajectory forecasting based on linear stationary models using automatic identification system,”Procedia Computer Sci- ence, vol. 218, pp. 1463–1474, 2023

    work page 2023

  14. [14]

    Short- term trajectory prediction of maritime vessel using k-nearest neighbor points,

    M. Zhang, L. Huang, Y . Wen, J. Zhang, Y . Huang, and M. Zhu, “Short- term trajectory prediction of maritime vessel using k-nearest neighbor points,”Journal of Marine Science and Engineering, vol. 10, no. 12, p. 1939, 2022

    work page 1939

  15. [15]

    Deep learning applications in vessel dead reckoning to deal with missing automatic identification system data,

    A. Sedaghat, H. Arbabkhah, M. Jafari Kang, and M. Hamidi, “Deep learning applications in vessel dead reckoning to deal with missing automatic identification system data,”Journal of Marine Science and Engineering, vol. 12, no. 1, p. 152, 2024

    work page 2024

  16. [16]

    Y . Wei, Z. Chen, C. Zhao, X. Chen, J. He, and C. Zhang, “A three-stage multi-objective heterogeneous integrated model with decomposition- reconstruction mechanism and adaptive segmentation error correction method for ship motion multi-step prediction,”Advanced Engineering Informatics, vol. 56, p. 101954, 2023

    work page 2023

  17. [17]

    Physics-informed machine learning models for ship speed prediction,

    X. Lang, D. Wu, and W. Mao, “Physics-informed machine learning models for ship speed prediction,”Expert Systems with Applications, vol. 238, p. 121877, 2024

    work page 2024

  18. [18]

    Math- data integrated prediction model for ship maneuvering motion,

    Q. Dong, N. Wang, J. Song, L. Hao, S. Liu, B. Han, and K. Qu, “Math- data integrated prediction model for ship maneuvering motion,”Ocean Engineering, vol. 285, p. 115255, 2023

    work page 2023

  19. [19]

    Ship trajectory prediction: An integrated approach using convlstm-based sequence-to-sequence model,

    W. Wu, P. Chen, L. Chen, and J. Mou, “Ship trajectory prediction: An integrated approach using convlstm-based sequence-to-sequence model,” Journal of Marine Science and Engineering, vol. 11, no. 8, p. 1484, 2023

    work page 2023

  20. [20]

    Automatic identification system (ais) data supported ship trajectory prediction and analysis via a deep learning model,

    X. Chen, C. Wei, G. Zhou, H. Wu, Z. Wang, and S. A. Biancardo, “Automatic identification system (ais) data supported ship trajectory prediction and analysis via a deep learning model,”Journal of Marine Science and Engineering, vol. 10, no. 9, p. 1314, 2022

    work page 2022

  21. [21]

    A ship trajectory prediction method based on gat and lstm,

    J. Zhao, Z. Yan, Z. Zhou, X. Chen, B. Wu, and S. Wang, “A ship trajectory prediction method based on gat and lstm,”Ocean engineering, vol. 289, p. 116159, 2023

    work page 2023

  22. [22]

    Toward multimodal vessel trajectory prediction by modeling the distribution of modes,

    S. Guo, H. Zhang, and Y . Guo, “Toward multimodal vessel trajectory prediction by modeling the distribution of modes,”Ocean Engineering, vol. 282, p. 115020, 2023

    work page 2023

  23. [23]

    A context-aware approach for vessels’ trajectory prediction,

    S. Mehri, A. A. Alesheikh, and A. Basiri, “A context-aware approach for vessels’ trajectory prediction,”Ocean Engineering, vol. 282, p. 114916, 2023

    work page 2023

  24. [24]

    An attention mechanism model based on positional encoding for the prediction of ship maneuvering motion in real sea state,

    L. Dong, H. Wang, and J. Lou, “An attention mechanism model based on positional encoding for the prediction of ship maneuvering motion in real sea state,”Journal of Marine Science and Technology, vol. 29, no. 1, pp. 136–152, 2024

    work page 2024

  25. [25]

    Large-scale long- term prediction of ship ais tracks via linear networks with a look-back window decomposition scheme of time features,

    W. Zhao, D. Wang, K. Gao, J. Wu, and X. Cheng, “Large-scale long- term prediction of ship ais tracks via linear networks with a look-back window decomposition scheme of time features,”Journal of Marine Science and Engineering, vol. 11, no. 11, p. 2132, 2023

    work page 2023

  26. [26]

    A transformer network with sparse aug- mented data representation and cross entropy loss for ais-based vessel trajectory prediction,

    D. Nguyen and R. Fablet, “A transformer network with sparse aug- mented data representation and cross entropy loss for ais-based vessel trajectory prediction,”IEEE Access, vol. 12, pp. 21 596–21 609, 2024

    work page 2024

  27. [27]

    Tripleconvtransformer: A deep learning vessel trajectory prediction method fusing discretized meteorological data,

    P. Huang, Q. Chen, D. Wang, M. Wang, X. Wu, and X. Huang, “Tripleconvtransformer: A deep learning vessel trajectory prediction method fusing discretized meteorological data,”Frontiers in Environ- mental Science, vol. 10, p. 1012547, 2022

    work page 2022

  28. [28]

    Informer: Beyond efficient transformer for long sequence time-series forecasting,

    H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, and W. Zhang, “Informer: Beyond efficient transformer for long sequence time-series forecasting,” inProceedings of the AAAI conference on artificial intel- ligence, vol. 35, no. 12, 2021, pp. 11 106–11 115

    work page 2021

  29. [29]

    A cnngru- mha method for ship trajectory prediction based on marine fusion data,

    J. Bi, M. Gao, K. Bao, W. Zhang, X. Zhang, and H. Cheng, “A cnngru- mha method for ship trajectory prediction based on marine fusion data,” Ocean Engineering, vol. 310, p. 118701, 2024

    work page 2024

  30. [30]

    Vessel trajectory prediction using vessel influence long short-term memory with uncertainty estimation,

    Z. Guo, H. Qiang, and X. Peng, “Vessel trajectory prediction using vessel influence long short-term memory with uncertainty estimation,”Journal of Marine Science and Engineering, vol. 13, no. 2, p. 353, 2025

    work page 2025

  31. [31]

    Spatio-temporal graph transformer networks for pedestrian trajectory prediction,

    C. Yu, X. Ma, J. Ren, H. Zhao, and S. Yi, “Spatio-temporal graph transformer networks for pedestrian trajectory prediction,” inComputer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XII 16. Springer, 2020, pp. 507–523

    work page 2020

  32. [32]

    What is a savitzky-golay filter?[lecture notes],

    R. W. Schafer, “What is a savitzky-golay filter?[lecture notes],”IEEE Signal processing magazine, vol. 28, no. 4, pp. 111–117, 2011

    work page 2011

  33. [33]

    L-vtp: Long- term vessel trajectory prediction based on multi-source data analysis,

    C. Liu, S. Guo, Y . Feng, F. Hong, H. Huang, and Z. Guo, “L-vtp: Long- term vessel trajectory prediction based on multi-source data analysis,” Sensors, vol. 19, no. 20, p. 4365, 2019

    work page 2019

  34. [34]

    Social lstm: Human trajectory prediction in crowded spaces,

    A. Alahi, K. Goel, V . Ramanathan, A. Robicquet, L. Fei-Fei, and S. Savarese, “Social lstm: Human trajectory prediction in crowded spaces,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 961–971

    work page 2016

  35. [35]

    Transformer networks for trajectory forecasting,

    F. Giuliari, I. Hasan, M. Cristani, and F. Galasso, “Transformer networks for trajectory forecasting,” in2020 25th international conference on pattern recognition (ICPR). IEEE, 2021, pp. 10 335–10 342. Ganeshaarajreceived the B.S. degree in Biomed- ical Engineering from the University of Moratuwa, Srilanka in 2022. He is currently pursuing a PhD de- gree ...

    work page 2021