Goal-based Trajectory Prediction for improved Cross-Dataset Generalization
This addresses the generalization issue for autonomous driving systems when deployed to new areas, though it appears incremental as it builds on existing GNN methods.
The paper tackles the problem of poor cross-dataset generalization in trajectory prediction for autonomous driving by introducing a Graph Neural Network that uses a heterogeneous graph with traffic participants and road networks to classify goals, resulting in improved performance when trained on Argoverse2 and evaluated on NuScenes.
To achieve full autonomous driving, a good understanding of the surrounding environment is necessary. Especially predicting the future states of other traffic participants imposes a non-trivial challenge. Current SotA-models already show promising results when trained on real datasets (e.g. Argoverse2, NuScenes). Problems arise when these models are deployed to new/unseen areas. Typically, performance drops significantly, indicating that the models lack generalization. In this work, we introduce a new Graph Neural Network (GNN) that utilizes a heterogeneous graph consisting of traffic participants and vectorized road network. Latter, is used to classify goals, i.e. endpoints of the predicted trajectories, in a multi-staged approach, leading to a better generalization to unseen scenarios. We show the effectiveness of the goal selection process via cross-dataset evaluation, i.e. training on Argoverse2 and evaluating on NuScenes.