Conditional Unscented Autoencoders for Trajectory Prediction

TL;DR

This paper introduces Conditional Unscented Autoencoders (CUAEs) for trajectory prediction, improving upon traditional CVAEs by using unscented sampling and structured latent spaces, leading to better performance on multiple datasets.

Contribution

The paper proposes a novel CUAEs framework that replaces random sampling with unscented sampling and enhances the latent space structure, significantly boosting trajectory prediction accuracy.

Findings

Outperforms state-of-the-art on INTERACTION dataset

Achieves better results on CelebA image modeling

Demonstrates the effectiveness of unscented sampling in CVAEs

Abstract

The CVAE is one of the most widely-used models in trajectory prediction for AD. It captures the interplay between a driving context and its ground-truth future into a probabilistic latent space and uses it to produce predictions. In this paper, we challenge key components of the CVAE. We leverage recent advances in the space of the VAE, the foundation of the CVAE, which show that a simple change in the sampling procedure can greatly benefit performance. We find that unscented sampling, which draws samples from any learned distribution in a deterministic manner, can naturally be better suited to trajectory prediction than potentially dangerous random sampling. We go further and offer additional improvements including a more structured Gaussian mixture latent space, as well as a novel, potentially more expressive way to do inference with CVAEs. We show wide applicability of our models by evaluating them on the INTERACTION prediction dataset, outperforming the state of the art, as well as at the task of image modeling on the CelebA dataset, outperforming the baseline vanilla CVAE. Code is available at https://github.com/boschresearch/cuae-prediction.