News

Latest research and engineering updates.

A lightweight timeline for papers, talks, releases, and long-term milestones.

VectorWorld was highlighted and endorsed in a keynote by the President of Bosch Intelligent Driving Control China.

Our VectorWorld work was specifically referenced and recognized in the invited keynote “Beneath the Cornerstone, Toward the Future: Automotive Intelligence and Global Expansion in the AI Era” delivered by Wu Yongqiao, President of Bosch Intelligent Driving Control (XC) China.

Coverage

VectorWorld was accepted to ICML as a Spotlight paper.

VectorWorld studies efficient streaming world models via diffusion flow on vector graphs.

Publication arXiv

RegFormer++ is available as an arXiv preprint.

RegFormer++ extends large-scale 3D LiDAR point registration with projection-aware 2D transformer design.

Publication arXiv

DifFlow3D was published in IEEE Transactions on Pattern Analysis and Machine Intelligence.

The journal version studies hierarchical diffusion models for uncertainty-aware 3D scene flow estimation.

Publication DOI

Unsupervised learning of 3D scene flow with LiDAR odometry assistance was published in IEEE T-ITS.

The work explores odometry-assisted unsupervised 3D scene flow learning.

Publication DOI

Mamba4D was published at CVPR 2025.

Mamba4D studies efficient long-sequence point cloud video understanding with disentangled spatial-temporal state space models.

Publication DOI

NeuroGauss4D-PCI was accepted to NeurIPS 2024.

The paper studies 4D neural fields and Gaussian deformation fields for point cloud interpolation.

Publication Code

NeuroGauss4D-PCI became available online.

A 4D point cloud interpolation framework based on neural fields and Gaussian deformation fields.

Publication

MAMBA4D became available online.

An efficient long-sequence point cloud video understanding model.

Publication

Invited talk by AutoDriver: CVPR 2024 work on 3DSFLabelling.

Invited by AutoDriver to share the CVPR 2024 work on 3D scene flow estimation and 3DSFLabelling.

Bilibili Talk Publication

3DSFLabelling was accepted to CVPR 2024.

The paper boosts 3D scene flow estimation by pseudo auto-labeling.

Publication Code

DifFlow3D was accepted to CVPR 2024.

The paper studies robust uncertainty-aware scene flow estimation with iterative diffusion-based refinement.

Publication DOI