top of page
Motion Prediction
Trends in Motion Prediction Toward Deployable and Generalizable Autonomy: A Revisit and Perspectives
Letian Wang, Marc-Antoine Lavoie, Sandro Papais, Barza Nisar, Yuxiao Chen, Wenhao Ding, Boris Ivanovic, Hao Shao, Abulikemu Abuduweili, Evan Cook, Yang Zhou, Peter Karkus, Jiachen Li, Changliu Liu, Marco Pavone, Steven Waslander
Foundation and Trends in Robotics, 2026.
This survey revisits the state of motion prediction, examining how intelligent systems like autonomous vehicles and robots can anticipate future agent states and scene evolution in dynamic, human-populated environments . While recent benchmark results have advanced rapidly, the paper identifies a critical gap between idealized research benchmarks and the complexity of real-world deployment, where state-of-the-art models often fail to generalize. To address this, the survey provides a comprehensive taxonomy of motion prediction methods, covering representations, modeling strategies, application domains, and evaluation protocols. It then focuses on two core challenges: making motion prediction models deployable within closed-loop autonomy stacks (integrating perception, prediction, planning, and control), and generalizing models from limited training scenarios to open-world conditions.
ForeSight: Multi-View Streaming Joint Object Detection and Trajectory Forecasting
Sandro Papais, Letian Wang, Brian Cheong, Steven L. Waslander
Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2025.
ForeSight is a novel joint detection and forecasting framework for vision-based 3D perception in autonomous vehicles that seamlessly integrates these tasks using a multi-task streaming and bidirectional learning architecture, allowing for shared query memory and improved temporal consistency.
SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction
Yang Zhou, Hao Shao, Letian Wang, Steven L. Waslander, Hongsheng Li, Yu Liu
International Conference on Learning Representations (ICLR), 2025.
SmartPretrain is a general and scalable self-supervised learning framework for motion prediction that can be applied across different models and datasets. By combining contrastive and reconstructive learning strategies, it captures complex spatiotemporal interactions without being restricted by model architecture. SmartPretrain introduces a dataset-agnostic scenario sampling strategy, leveraging diverse datasets to improve robustness and generalization.
SmartRefine: A Scenario-Adaptive Refinement Framework for Efficient Motion Prediction
Yang Zhou, Hao Shao, Letian Wang, Steven L. Waslander, Hongsheng Li, Yu Liu
Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR), 2024
We introduce SmartRefine, a novel scenario-adaptive refinement strategy, which can comprehensively adapt refinement configurations based on each scenario's properties. It intelligently selects the number of refinement iterations by introducing a quality score that measures the prediction quality and remaining refinement potential of each scenario.
bottom of page