12 April 2019 in IEEE Sensors Journal
Multi-sensor fusion plays a key role in 2D laser based robot location and navigation. Although it has achieved great success, there are still some challenges, e.g., being fragile when having large angular rotation. In this paper, we present a deep learning based approach to localizing a mobile robot using a 2D laser and an Inertial Measurement Unit. A novel Recurrent Convolutional Neural Network (RCNN) based architecture is developed to fuse laser and inertial data for scan-to-scan pose estimation. A scan-to-submap optimization is also introduced to optimize the poses estimated by the RCNN for enhanced robustness and accuracy. Extensive experiments have been conducted in both simulation and practice with a real mobile robot, verifying the effectiveness of the proposed deep sensor fusion system.
2 August 2022 in Transactions of the Institute of Measurement and Control
Light detection and ranging (LiDAR) odometry plays a crucial role in autonomous mobile robots and unmanned ground vehicles (UGVs). This paper presents a deep learning–based odometry system using two successive three-dimensional (3D) point clouds to estimate their scene flow and then predict their relative pose. The network consumes continuous 3D point clouds directly and outputs their scene flow and uncertain mask in a coarse-to-fine fashion. A pose estimation layer without trainable parameters is designed to compute the pose with the scene flow. We also introduce a scan-to-map optimization algorithm to enhance the robustness and accuracy of the system. Our experiments on the KITTI odometry data set and our campus data set demonstrate the effectiveness of the proposed deep learning–based point cloud odometry.
21 July 2022 in IET Cyber-Systems and Robotics
Deep‐learning‐based 3D place recognition has received more attention since the data‐driven fashion is widely used for the 3D point cloud applications. Most of the existing deep‐learning‐based 3D place recognition methods only utilise a single scene for place recognition. However, a single scene may have measurement noise or observable dynamic object differences, which may lead to a reduction in recognition accuracy. To improve the performance of 3D place recognition, a sequence matching based rearrangement method is proposed. Our sequence matching method is based on an assignment algorithm and guides the candidate rearrangement in searching for a similar place. The global descriptor extraction adapts the effective sparse tensor representation and a simple pooling layer to obtain the global descriptor. A new loss function combination is employed to train the network. The proposed approach is evaluated on the popular 3D place recognition benchmarks, which proves the effectiveness of the proposed approach.
1 January 2026 in IET Cyber-Systems and Robotics
Real‐time dense three‐dimensional (3D) semantic mapping plays a crucial role in robotics. However, existing explicit and implicit representations often entail a trade‐off among geometric precision, semantic richness and computational efficiency. To address this challenge, we propose a hierarchical hybrid mapping framework that unifies geometry, appearance and dense semantics within a shared multiresolution hash grid, enabling end‐to‐end optimisation via differentiable rendering. By caching per‐keyframe semantic masks and reusing them during incremental refinement, our method eliminates redundant semantic segmentation and significantly accelerates runtime. Additionally, an adaptive keyframe pruning strategy further ensures bounded memory usage. Extensive experiments demonstrate that our framework achieves real‐time performance with superior geometric and semantic accuracy.