Advanced Research: Nullmax's Perception Paper Accepted by RA-L

Nullmax Perception Team's research paper, "Multi-Correlation Siamese Transformer Network with Dense Connection for 3D Single Object Tracking," was accepted at IEEE Robotics and Automation Letters (RA-L). This recognition reaffirms Nullmax's R&D prowess.

The paper introduces an innovative Transformer Siamese network, demonstrating excellent performance and generalization in sparse point cloud environments. This advancement enhances Nullmax's multi-modal perception capabilities, vital for the future of autonomous driving.

Paper link: arxiv.org/abs/2312.11051

Code repository: github.com/liangp/MCSTN-3DSOT

Main Content of the Paper

The proposed network incorporates multiple stages of feature correlation using sparse pillar features, applying self-attention and cross-attention mechanisms to enhance 3D single-object tracking. This approach significantly improves the performance and optimization of sparse point cloud-based tracking algorithms, with dense connections facilitating efficient learning and prediction of target states.

Main contributions

The team proposes to inject the intact template information intothe search area at multiple points of the network so thatthe feature extraction of the search area is aware of thetemplate for Siamese-based 3D SOT.

Introduced a multi-stage densely connected SiameseTransformer network that uses self-attention for separatefeature learning of each branch and cross-attention forfeature correlation between the template and search areabased on sparse pillars in each stage.

Comprehensive experiments on the KITTI, nuScenes, and Waymo show the performanceof the proposed algorithm is promising in comparisonwith the state-of-the-art algorithms.

As a leading automated driving company, Nullmax has advanced through significant development stages, delivering competitive and mass-produced autonomous driving solutions. Entering its 3.0 phase, Nullmax continues to accelerate mass production while investing in next-generation technology development, striving for higher levels of autonomous driving.