SHARP: Short-Window Streaming for Accurate and Robust Prediction in Motion Forecasting

TL;DR

This paper introduces SHARP, a streaming motion forecasting framework that maintains high accuracy and robustness across varying observation lengths in dynamic traffic scenes, suitable for real-time applications.

Contribution

The authors propose a novel streaming-based motion forecasting method with instance-aware context streaming and dual training, improving robustness across diverse observation horizons.

Findings

Achieves state-of-the-art streaming inference performance on Argoverse 2 benchmark.

Demonstrates robustness under evolving scene conditions and on single-agent benchmarks.

Maintains minimal latency suitable for real-world deployment.

Abstract

In dynamic traffic environments, motion forecasting models must be able to accurately estimate future trajectories continuously. Streaming-based methods are a promising solution, but despite recent advances, their performance often degrades when exposed to heterogeneous observation lengths. To address this, we propose a novel streaming-based motion forecasting framework that explicitly focuses on evolving scenes. Our method incrementally processes incoming observation windows and leverages an instance-aware context streaming to maintain and update latent agent representations across inference steps. A dual training objective further enables consistent forecasting accuracy across diverse observation horizons. Extensive experiments on Argoverse 2, nuScenes, and Argoverse 1 demonstrate the robustness of our approach under evolving scene conditions and also on the single-agent benchmarks. Our model achieves state-of-the-art performance in streaming inference on the Argoverse 2 multi-agent benchmark, while maintaining minimal latency, highlighting its suitability for real-world deployment.