Problem
Research questions and friction points this paper is trying to address.
motion forecasting
streaming inference
heterogeneous observation lengths
dynamic traffic environments
trajectory prediction
SHARP: Short-Window Streaming for Accurate and Robust Prediction in Motion Forecasting
🤖 AI Summary
This work addresses the performance degradation in streaming motion prediction under dynamic traffic scenarios caused by heterogeneous observation window lengths. To this end, the authors propose an evolution-aware streaming prediction framework that leverages short-window incremental inference, instance-aware contextual flow, and continuous updating of implicit representations. A dual training objective is introduced to ensure prediction consistency across varying observation durations. Evaluated on the Argoverse 2 multi-agent streaming prediction benchmark, the proposed method achieves state-of-the-art performance and demonstrates strong results on Argoverse 1 and nuScenes as well. The framework exhibits high robustness, accuracy, and low latency, making it well-suited 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.
Innovation
Methods, ideas, or system contributions that make the work stand out.
streaming motion forecasting
instance-aware context streaming
short-window processing
dual training objective
robust trajectory prediction
🔎 Similar Papers
No similar papers found.
