Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GOHOME: Graph-Oriented Heatmap Output for future Motion Estimation

Thomas Gilles, Stefano Sabatini, Dzmitry Tsishkou, Bogdan Stanciulescu, Fabien Moutarde

TL;DR

GOHOME introduces a graph-oriented heatmap approach for future motion estimation by encoding HD-Map lanelets with a graph neural network and generating lane-level rasters that project into a global heatmap of future positions. This design avoids full-image CNNs, enables multimodal and uncertainty-aware predictions, and uses sparse sampling plus ranking to produce final trajectories efficiently. The method achieves competitive or state-of-the-art results across Argoverse, NuScenes, and Interaction datasets, with substantial speed-ups and memory reductions compared to CNN-based baselines, and benefits further from model ensembling. Overall, GOHOME demonstrates scalable, accurate, and versatile trajectory prediction suitable for real-time autonomous navigation and decision-making.

Abstract

In this paper, we propose GOHOME, a method leveraging graph representations of the High Definition Map and sparse projections to generate a heatmap output representing the future position probability distribution for a given agent in a traffic scene. This heatmap output yields an unconstrained 2D grid representation of agent future possible locations, allowing inherent multimodality and a measure of the uncertainty of the prediction. Our graph-oriented model avoids the high computation burden of representing the surrounding context as squared images and processing it with classical CNNs, but focuses instead only on the most probable lanes where the agent could end up in the immediate future. GOHOME reaches 2$nd$ on Argoverse Motion Forecasting Benchmark on the MissRate$_6$ metric while achieving significant speed-up and memory burden diminution compared to Argoverse 1$^{st}$ place method HOME. We also highlight that heatmap output enables multimodal ensembling and improve 1$^{st}$ place MissRate$_6$ by more than 15$\%$ with our best ensemble on Argoverse. Finally, we evaluate and reach state-of-the-art performance on the other trajectory prediction datasets nuScenes and Interaction, demonstrating the generalizability of our method.

GOHOME: Graph-Oriented Heatmap Output for future Motion Estimation

TL;DR

GOHOME introduces a graph-oriented heatmap approach for future motion estimation by encoding HD-Map lanelets with a graph neural network and generating lane-level rasters that project into a global heatmap of future positions. This design avoids full-image CNNs, enables multimodal and uncertainty-aware predictions, and uses sparse sampling plus ranking to produce final trajectories efficiently. The method achieves competitive or state-of-the-art results across Argoverse, NuScenes, and Interaction datasets, with substantial speed-ups and memory reductions compared to CNN-based baselines, and benefits further from model ensembling. Overall, GOHOME demonstrates scalable, accurate, and versatile trajectory prediction suitable for real-time autonomous navigation and decision-making.

Abstract

In this paper, we propose GOHOME, a method leveraging graph representations of the High Definition Map and sparse projections to generate a heatmap output representing the future position probability distribution for a given agent in a traffic scene. This heatmap output yields an unconstrained 2D grid representation of agent future possible locations, allowing inherent multimodality and a measure of the uncertainty of the prediction. Our graph-oriented model avoids the high computation burden of representing the surrounding context as squared images and processing it with classical CNNs, but focuses instead only on the most probable lanes where the agent could end up in the immediate future. GOHOME reaches 2 on Argoverse Motion Forecasting Benchmark on the MissRate metric while achieving significant speed-up and memory burden diminution compared to Argoverse 1 place method HOME. We also highlight that heatmap output enables multimodal ensembling and improve 1 place MissRate by more than 15 with our best ensemble on Argoverse. Finally, we evaluate and reach state-of-the-art performance on the other trajectory prediction datasets nuScenes and Interaction, demonstrating the generalizability of our method.

Paper Structure

This paper contains 23 sections, 2 equations, 6 figures, 5 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. METHOD
  4. Graph neural network for HD-Map input
  5. Heatmap generation through Lane-level rasters
  6. Lane raster generation
  7. Lane ranking
  8. Cartesian image connection
  9. Sparse sampling for Miss Rate Optimization and Full trajectory generation
  10. Model ensembling
  11. Experimental Results
  12. Experimental settings
  13. Datasets
  14. Metrics
  15. Implementation details
  16. ...and 8 more sections

Figures (6)

  • Figure 1: GOHOME pipeline. The lane graph extracted from the HD-Map is processed through a graph encoder. Each lane then generates a local curvilinear raster that is combined into a predicted probability distribution heatmap.
  • Figure 2: GOHOME model architecture
  • Figure 3: Lane raster grid projection onto cartesian coordinates. a) A single node of the graph is a lanelet and describes a road segment. b) A rectangular raster is generated along the curvilinear coordinates of the lanelet. c) The lanelet coordinates are then used to project the predicted raster back into cartesian coordinates to complete the final heatmap output.
  • Figure 4: Inference time with regard to output range
  • Figure 5: Inference time with regard to pixels per meters
  • ...and 1 more figures