TopoDiffuser: A Diffusion-Based Multimodal Trajectory Prediction Model with Topometric Maps

TL;DR

TopoDiffuser is a novel diffusion-based model that leverages topometric maps for accurate, diverse, and road-compliant vehicle trajectory prediction, outperforming existing methods on the KITTI benchmark.

Contribution

It introduces a diffusion framework that embeds topometric map cues into trajectory prediction, enabling natural adherence to road geometry without explicit constraints.

Findings

Outperforms state-of-the-art on KITTI benchmark

Maintains strong geometric consistency

Ablation studies confirm modality contributions

Abstract

This paper introduces TopoDiffuser, a diffusion-based framework for multimodal trajectory prediction that incorporates topometric maps to generate accurate, diverse, and road-compliant future motion forecasts. By embedding structural cues from topometric maps into the denoising process of a conditional diffusion model, the proposed approach enables trajectory generation that naturally adheres to road geometry without relying on explicit constraints. A multimodal conditioning encoder fuses LiDAR observations, historical motion, and route information into a unified bird's-eye-view (BEV) representation. Extensive experiments on the KITTI benchmark demonstrate that TopoDiffuser outperforms state-of-the-art methods, while maintaining strong geometric consistency. Ablation studies further validate the contribution of each input modality, as well as the impact of denoising steps and the number of trajectory samples. To support future research, we publicly release our code at https://github.com/EI-Nav/TopoDiffuser.