RMP: A Random Mask Pretrain Framework for Motion Prediction

TL;DR

This paper introduces RMP, a pretraining framework for motion prediction in autonomous driving that masks object positions at random times and learns to fill them in, improving accuracy especially in occlusion scenarios.

Contribution

The paper proposes a novel pretraining framework inspired by NLP and CV masked models, tailored for trajectory prediction in autonomous driving.

Findings

Improves motion prediction accuracy on Argoverse and NuScenes datasets.

Enhances robustness to noisy inputs and occlusions.

Flexible framework that can adapt to various motion-related tasks.

Abstract

As the pretraining technique is growing in popularity, little work has been done on pretrained learning-based motion prediction methods in autonomous driving. In this paper, we propose a framework to formalize the pretraining task for trajectory prediction of traffic participants. Within our framework, inspired by the random masked model in natural language processing (NLP) and computer vision (CV), objects' positions at random timesteps are masked and then filled in by the learned neural network (NN). By changing the mask profile, our framework can easily switch among a range of motion-related tasks. We show that our proposed pretraining framework is able to deal with noisy inputs and improves the motion prediction accuracy and miss rate, especially for objects occluded over time by evaluating it on Argoverse and NuScenes datasets.