CrossLiDAR: Controllable Multi-Sensor LiDAR Sequence Generation
CrossLiDAR: Controllable Multi-Sensor LiDAR Sequence Generation
This paper introduces CrossLiDAR, a unified transformer-based framework designed to generate high-fidelity, controllable, and spatio-temporally consistent multi-sensor LiDAR sequences. The system utilizes a rectified flow objective with a Multi-Modal Diffusion Transformer (MM-DiT) backbone within a compact latent space of range images. Key components include an Inter-sensor Module for cross-sensor interaction and a ControlNet-style branch for structural conditioning using BEV priors or 3D bounding boxes. To ensure long-horizon coherence, the model generates short N-frame sequences autoregressively. Evaluated on the nuScenes and Waymo datasets, CrossLiDAR achieves an average improvement of 29.56% in JSD and 15.36% in MMD compared to state-of-the-art baselines, addressing bottlenecks in scalable autonomous system development.
The authors present CrossLiDAR, a transformer-based framework for generating multi-sensor LiDAR sequences with controllable geometry and temporal consistency, evaluated on nuScenes and Waymo datasets.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
er-based framework capable of generating controllable and spatio-temporally consistent multi-sensor LiDAR sequences. Our approach leverages a rectified flow objective with a Multi-Modal Diffusion Transformer (MM-DiT) backbone, within a compact latent space of range images. To support multiple LiDAR sensors, an Inter-sensor Module enables interaction across sensors, while a ControlNet-style branch provides structural
Open source S3
conditioning (e.g., BEV priors or 3D bounding boxes). For modeling the temporal consistency, CrossLiDAR generates short $N$-frame sequences and extends them autoregressively to improve long-horizon coherence. Our model significantly outperforms state-of-the-art baselines, achieving an average improvement of 29.56\% in JSD and 15.36\% in MMD evaluated on nuScenes and the Waymo Dataset. CrossLiDAR bridges the gap to mo
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org