Likelihood-Based Diverse Sampling for Trajectory Forecasting

TL;DR

This paper introduces Likelihood-Based Diverse Sampling (LDS), a novel method to generate diverse and high-likelihood trajectory samples from pre-trained flow models, improving multi-modal forecasting for vehicles and pedestrians.

Contribution

LDS is a new sampling technique that enhances diversity and likelihood of trajectory predictions, applicable to various flow models and capable of on-the-fly training for unseen data.

Findings

LDS outperforms existing diverse forecasting methods on multiple benchmarks.

LDS can be integrated as a plug-in to improve pre-trained models.

LDS enables transductive forecasting with on-the-fly training.

Abstract

Forecasting complex vehicle and pedestrian multi-modal distributions requires powerful probabilistic approaches. Normalizing flows (NF) have recently emerged as an attractive tool to model such distributions. However, a key drawback is that independent samples drawn from a flow model often do not adequately capture all the modes in the underlying distribution. We propose Likelihood-Based Diverse Sampling (LDS), a method for improving the quality and the diversity of trajectory samples from a pre-trained flow model. Rather than producing individual samples, LDS produces a set of trajectories in one shot. Given a pre-trained forecasting flow model, we train LDS using gradients from the model, to optimize an objective function that rewards high likelihood for individual trajectories in the predicted set, together with high spatial separation among trajectories. LDS outperforms state-of-art post-hoc neural diverse forecasting methods for various pre-trained flow models as well as conditional variational autoencoder (CVAE) models. Crucially, it can also be used for transductive trajectory forecasting, where the diverse forecasts are trained on-the-fly on unlabeled test examples. LDS is easy to implement, and we show that it offers a simple plug-in improvement over baselines on two challenging benchmarks. Code is at: https://github.com/JasonMa2016/LDS