Likely, Light, and Accurate Context-Free Clusters-based Trajectory Prediction

TL;DR

This paper introduces a multi-stage probabilistic trajectory forecasting system using a novel deep clustering method and distance-based ranking, outperforming existing context-free models in accuracy and efficiency.

Contribution

It presents a new deep feature clustering technique with self-conditioned GANs and a novel ranking method for trajectory prediction, improving robustness and computational efficiency.

Findings

Outperforms context-free deep generative models in trajectory accuracy.

Achieves similar performance to point estimators for top trajectories.

Demonstrates robustness to distribution shifts.

Abstract

Autonomous systems in the road transportation network require intelligent mechanisms that cope with uncertainty to foresee the future. In this paper, we propose a multi-stage probabilistic approach for trajectory forecasting: trajectory transformation to displacement space, clustering of displacement time series, trajectory proposals, and ranking proposals. We introduce a new deep feature clustering method, underlying self-conditioned GAN, which copes better with distribution shifts than traditional methods. Additionally, we propose novel distance-based ranking proposals to assign probabilities to the generated trajectories that are more efficient yet accurate than an auxiliary neural network. The overall system surpasses context-free deep generative models in human and road agents trajectory data while performing similarly to point estimators when comparing the most probable trajectory.