Forecasting the Past: Gradient-Based Distribution Shift Detection in Trajectory Prediction

arXiv cs.LG / 4/15/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper addresses the problem that trajectory prediction models degrade in automated driving when training and test conditions differ due to distributional shifts.
It introduces a self-supervised, post-hoc method that learns a decoder to forecast the second half of observed trajectories and uses the L2 norm of the gradient (w.r.t. the decoder’s final layer) as a distribution-shift score.
The approach is designed to avoid interfering with the original trajectory prediction model by training only the additional decoder in a separate step.
Experiments report substantial improvements in distribution shift detection on the Shifts and Argoverse datasets.
The method is also demonstrated as an early-warning mechanism for collisions in a deep Q-network motion planner within the Highway simulator.

Abstract

Trajectory prediction models often fail in real-world automated driving due to distributional shifts between training and test conditions. Such distributional shifts, whether behavioural or environmental, pose a critical risk by causing the model to make incorrect forecasts in unfamiliar situations. We propose a self-supervised method that trains a decoder in a post-hoc fashion on the self-supervised task of forecasting the second half of observed trajectories from the first half. The L2 norm of the gradient of this forecasting loss with respect to the decoder's final layer defines a score to identify distribution shifts. Our approach, first, does not affect the trajectory prediction model, ensuring no interference with original prediction performance and second, demonstrates substantial improvements on distribution shift detection for trajectory prediction on the Shifts and Argoverse datasets. Moreover, we show that this method can also be used to early detect collisions of a deep Q-Network motion planner in the Highway simulator. Source code is available at https://github.com/Michedev/forecasting-the-past.