Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Navigation under uncertainty: Trajectory prediction and occlusion reasoning with switching dynamical systems

Ran Wei, Joseph Lee, Shohei Wakayama, Alexander Tschantz, Conor Heins, Christopher Buckley, John Carenbauer, Hari Thiruvengada, Mahault Albarracin, Miguel de Prado, Petter Horling, Peter Winzell, Renjith Rajagopal

TL;DR

This work exploring a conceptual framework unifying trajectory prediction and occlusion reasoning under the same class of structured probabilistic generative models, namely, switching dynamical systems, shows that these models can achieve both higher prediction accuracy and uncertainty calibration compared to conditional Gaussian mixture models that are widely used for trajectory prediction.

Abstract

Predicting future trajectories of nearby objects, especially under occlusion, is a crucial task in autonomous driving and safe robot navigation. Prior works typically neglect to maintain uncertainty about occluded objects and only predict trajectories of observed objects using high-capacity models such as Transformers trained on large datasets. While these approaches are effective in standard scenarios, they can struggle to generalize to the long-tail, safety-critical scenarios. In this work, we explore a conceptual framework unifying trajectory prediction and occlusion reasoning under the same class of structured probabilistic generative model, namely, switching dynamical systems. We then present some initial experiments illustrating its capabilities using the Waymo open dataset.

Navigation under uncertainty: Trajectory prediction and occlusion reasoning with switching dynamical systems

TL;DR

This work exploring a conceptual framework unifying trajectory prediction and occlusion reasoning under the same class of structured probabilistic generative models, namely, switching dynamical systems, shows that these models can achieve both higher prediction accuracy and uncertainty calibration compared to conditional Gaussian mixture models that are widely used for trajectory prediction.

Abstract

Predicting future trajectories of nearby objects, especially under occlusion, is a crucial task in autonomous driving and safe robot navigation. Prior works typically neglect to maintain uncertainty about occluded objects and only predict trajectories of observed objects using high-capacity models such as Transformers trained on large datasets. While these approaches are effective in standard scenarios, they can struggle to generalize to the long-tail, safety-critical scenarios. In this work, we explore a conceptual framework unifying trajectory prediction and occlusion reasoning under the same class of structured probabilistic generative model, namely, switching dynamical systems. We then present some initial experiments illustrating its capabilities using the Waymo open dataset.

Paper Structure

This paper contains 18 sections, 9 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Traffic scene modeling with switching dynamical systems
  3. Trajectory prediction with switching dynamical systems
  4. Occlusion reasoning and object tracking with switching dynamical systems
  5. Single object tracking
  6. Multi object tracking
  7. Complete model
  8. Inference
  9. Experiments
  10. Data processing
  11. Baseline models and ablations
  12. Evaluation metrics
  13. Trajectory prediction results
  14. Occlusion reasoning results
  15. Conclusion and Future Work
  16. ...and 3 more sections

Figures (5)

  • Figure 1: Visualizations of predicted vehicle trajectories.
  • Figure 2: Visualizations of predicted vehicle trajectories by context length.
  • Figure 3: Visualization of occlusion inference. The beige-colored polygons in all plots represent the occluded regions at $t=0$. The blue-colored polygons represent the occluded regions at the time step of the respective plot. Vehicles are represented by gray boxes, and the ego vehicle is highlighted by a green star. Pedestrians are represented by crosses where observed ones at $t=0$ are colored in green and occluded ones are colored in red. The circular contours represent the Gaussian belief over pedestrian positions, and their transparency represents the model's beliefs over the existence of the corresponding pedestrian.
  • Figure 4: Visualizations of predicted pedestrian trajectories.
  • Figure 5: Visualizations of predicted pedestrian trajectories by context length.