Efficient Data Representation for Motion Forecasting: A Scene-Specific Trajectory Set Approach

TL;DR

This work tackles the challenge of representing diverse, plausible futures for motion forecasting by introducing scene-specific trajectory sets that leverage HD-map information and actor dynamics. It couples a deterministic goal sampler with Recursive In-Distribution Subsampling (RIDS) to generate tailored trajectory sets, balancing diversity and plausibility. On Argoverse 2 data, the approach achieves up to a 10% improvement in Driving Area Compliance (DAC) while maintaining competitive displacement errors, demonstrating that scene-aware representations better capture real-world heterogeneity. The method offers practical benefits for data-efficient, map-consistent forecasting in autonomous driving pipelines.

Abstract

Representing diverse and plausible future trajectories is critical for motion forecasting in autonomous driving. However, efficiently capturing these trajectories in a compact set remains challenging. This study introduces a novel approach for generating scene-specific trajectory sets tailored to different contexts, such as intersections and straight roads, by leveraging map information and actor dynamics. A deterministic goal sampling algorithm identifies relevant map regions, while our Recursive In-Distribution Subsampling (RIDS) method enhances trajectory plausibility by condensing redundant representations. Experiments on the Argoverse 2 dataset demonstrate that our method achieves up to a 10% improvement in Driving Area Compliance (DAC) compared to baseline methods while maintaining competitive displacement errors. Our work highlights the benefits of mining such scene-aware trajectory sets and how they could capture the complex and heterogeneous nature of actor behavior in real-world driving scenarios.

Figures (3)

• Figure 1: A Set for (a) Intersection Scenes and (b) Non-Intersection Scenes
• Figure 2: Steps involved in clustering the scenes based on the road geometry and actor states are shown here. Past trajectory and Future trajectory of an FT are illustrated. In (a), Grey connections denote the connectivity between various lanes at an intersection. In (b), Blue edges depict the transformation of lane connections into node edges. Using \ref{['Alg:alg1']}, the Source node, which corresponds to the closest node relative to the last observed state of the FT, is identified. Finally, the edges in (c) represent the plausible lanes the FT could potentially select.
• Figure 3: The randomly sampled sets can retain more number of trajectories when compared with a metric-driven set, making the former a reliable way to represent more plausible states