mlsanityData to 7 Oct 2026

TaPD: Temporal-adaptive Progressive Distillation for Observation-Adaptive Trajectory Forecasting in Autonomous Driving

arXiv 2603.06231 ↗First posted 6 Mar 2026cs.CV cs.AI cs.ROcited by 0

Not on the map. Why?

Abstract (arXiv)

Trajectory prediction is essential for autonomous driving, enabling vehicles to anticipate the motion of surrounding agents to support safe planning. However, most existing predictors assume fixed-length histories and suffer substantial performance degradation when observations are variable or extremely short in real-world settings (e.g., due to occlusion or a limited sensing range). We propose TaPD (Temporal-adaptive Progressive Distillation), a unified plug-and-play framework for observation-adaptive trajectory forecasting under variable history lengths. TaPD comprises two cooperative modules: an Observation-Adaptive Forecaster (OAF) for future prediction and a Temporal Backfilling Module (TBM) for explicit reconstruction of the past. OAF is built on progressive knowledge distillation (PKD), which transfers motion pattern knowledge from long-horizon "teachers" to short-horizon "students" via hierarchical feature regression, enabling short observations to recover richer motion context. We further introduce a cosine-annealed distillation weighting scheme to balance forecasting supervision and feature alignment, improving optimization stability and cross-length consistency. For extremely short histories where implicit alignment is insufficient, TBM backfills missing historical segments conditioned on scene evolution, producing context-rich trajectories that strengthen PKD and thereby improve OAF. We employ a decoupled pretrain-reconstruct-finetune protocol to preserve real-motion priors while adapting to backfilled inputs. Extensive experiments on Argoverse 1 and Argoverse 2 show that TaPD consistently outperforms strong baselines across all observation lengths, delivers especially large gains under very short inputs, and improves other predictors (e.g., HiVT) in a plug-and-play manner. Code will be available at https://github.com/zhouhao94/TaPD.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 52 of its 70 references that are in the map. First posted on arXiv 6 Mar 2026.

Builds on

52 of 70 references.

References · 52

  1. 2017Attention Is All You Needcited by 59,932
  2. 2019Scalability in Perception for Autonomous Driving: Waymo Open Datasetcited by 2,436
  3. 2019Argoverse: 3D Tracking and Forecasting with Rich Mapscited by 728
  4. 2023Argoverse 2: Next Generation Datasets for Self-Driving Perception and Forecastingcited by 701
  5. 2020VectorNet: Encoding HD Maps and Agent Dynamics from Vectorized Representationcited by 528
  6. 2019MultiPath: Multiple Probabilistic Anchor Trajectory Hypotheses for Behavior Predictioncited by 417
  7. 2020Learning Lane Graph Representations for Motion Forecastingcited by 380
  8. 2019A Survey of Autonomous Driving: Common Practices and Emerging Technologiescited by 373
  9. 2020TNT: Target-driveN Trajectory Predictioncited by 345
  10. 2021DenseTNT: End-to-end Trajectory Prediction from Dense Goal Setscited by 292
  11. 2022Wayformer: Motion Forecasting via Simple & Efficient Attention Networkscited by 279
  12. 2022Motion Transformer with Global Intention Localization and Local Movement Refinementcited by 267
38 more
  1. 2019CoverNet: Multimodal Behavior Prediction using Trajectory Setscited by 259
  2. 2023MotionDiffuser: Controllable Multi-Agent Motion Prediction using Diffusioncited by 210
  3. 2021Multimodal Motion Prediction with Stacked Transformerscited by 207
  4. 2021HOME: Heatmap Output for future Motion Estimationcited by 122
  5. 2021LaneRCNN: Distributed Representations for Graph-Centric Motion Forecastingcited by 112
  6. 2021TPCN: Temporal Point Cloud Networks for Motion Forecastingcited by 94
  7. 2023Forecast-MAE: Self-supervised Pre-training for Motion Forecasting with Masked Autoencoderscited by 72
  8. 2022GANet: Goal Area Network for Motion Forecastingcited by 49
  9. 2023ProphNet: Efficient Agent-Centric Motion Forecasting with Anchor-Informed Proposalscited by 47
  10. 2023Uncovering the Missing Pattern: Unified Framework Towards Trajectory Imputation and Predictioncited by 47
  11. 2023Leveraging Future Relationship Reasoning for Vehicle Trajectory Predictioncited by 47
  12. 2023Traj-MAE: Masked Autoencoders for Trajectory Predictioncited by 42
  13. 2023ADAPT: Efficient Multi-Agent Trajectory Prediction with Adaptationcited by 39
  14. 2024HPNet: Dynamic Trajectory Forecasting with Historical Prediction Attentioncited by 37
  15. 2022How many Observations are Enough? Knowledge Distillation for Trajectory Forecastingcited by 35
  16. 2024SmartRefine: A Scenario-Adaptive Refinement Framework for Efficient Motion Predictioncited by 33
  17. 2022SSL-Lanes: Self-Supervised Learning for Motion Forecasting in Autonomous Drivingcited by 31
  18. 2023SEPT: Towards Efficient Scene Representation Learning for Motion Predictioncited by 27
  19. 2024SIMPL: A Simple and Efficient Multi-agent Motion Prediction Baseline for Autonomous Drivingcited by 27
  20. 2022Path-Aware Graph Attention for HD Maps in Motion Predictioncited by 23
  21. 2022R-Pred: Two-Stage Motion Prediction Via Tube-Query Attention-Based Trajectory Refinementcited by 23
  22. 2024Adapting to Length Shift: FlexiLength Network for Trajectory Predictioncited by 22
  23. 2024T4P: Test-Time Training of Trajectory Prediction via Masked Autoencoder and Actor-specific Token Memorycited by 20
  24. 2021Trajectory Prediction with Graph-based Dual-scale Context Fusioncited by 19
  25. 2025Trajectory Mamba: Efficient Attention-Mamba Forecasting Model Based on Selective SSMcited by 17
  26. 2023Scene Informer: Anchor-based Occlusion Inference and Trajectory Prediction in Partially Observable Environmentscited by 14
  27. 2025Trajectory Prediction for Autonomous Driving: Progress, Limitations, and Future Directionscited by 14
  28. 2024Motion Forecasting in Continuous Drivingcited by 11
  29. 2025LANGTRAJ: Diffusion Model and Dataset for Language-Conditioned Trajectory Simulationcited by 11
  30. 2023Attention-aware Social Graph Transformer Networks for Stochastic Trajectory Predictioncited by 8
  31. 2024SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Predictioncited by 7
  32. 2024Quantifying Uncertainty in Motion Prediction with Variational Bayesian Mixturecited by 6
  33. 2024FutureNet-LOF: Joint Trajectory Prediction and Lane Occupancy Field Prediction with Future Context Encodingcited by 6
  34. 2025DONUT: A Decoder-Only Model for Trajectory Predictioncited by 6
  35. 2025Generative Active Learning for Long-tail Trajectory Prediction via Controllable Diffusion Modelcited by 6
  36. 2025Relative Position Matters: Trajectory Prediction and Planning with Polar Representationcited by 5
  37. 2025Future-Aware Interaction Network For Motion Forecastingcited by 4
  38. 2024A Structure-Aware Lane Graph Transformer Model for Vehicle Trajectory Predictioncited by 2

50 most cited of 52 All 52 (JSON)

Built on by

Not cited yet.