Micro air vehicles (MAVs) enable high-speed autonomous navigation for search and rescue but face challenges in weight, sensing, and control. We present SUPER, a 280-mm MAV with a thrust-to-weight ratio >5.0 for agile flight in cluttered environments. It uses a lightweight 3D LiDAR for long-range obstacle detection and an efficient planning framework that generates two trajectories—one for safety, another for speed. Compared to baselines, SUPER reduced failure rates by 35.9x, halved planning time, and achieved 20 m/s speeds, successfully avoiding obstacles. SUPER marks a milestone in MAV autonomy, bridging lab research to real-world applications.

Authors: Yunfan Ren, Fangcheng Zhu, Guozheng Lu, Yixi Cai, Longji Yin, Fanze Kong, Jiarong Lin, Nan Chen, Fu Zhang
Videos: bilibili   youtube   Code: Github   Star   Paper: Link

Fast-LIVO2 is a fast, direct LiDAR-inertial-visual odometry framework designed for accurate and robust state estimation in SLAM tasks, ideal for real-time, onboard robotic applications. It fuses IMU, LiDAR, and image measurements using an error-state iterated Kalman filter (ESIKF) and a sequential update strategy. Direct methods are employed for visual and LiDAR fusion, using a unified voxel map for geometric structure and image alignment. Plane priors and dynamic reference patch updates enhance image alignment accuracy. Extensive experiments show Fast-LIVO2's superior accuracy, robustness, and computational efficiency. Applications include UAV navigation, airborne mapping, and 3D model rendering.

Authors: Chunran Zheng, Wei Xu, Zuhao Zou, Tong Hua, Chongjian Yuan, Dongjiao He, Bingyang Zhou, Zheng Liu, Jiarong Lin, Fangcheng Zhu, Yunfan Ren, Rong Wang, Fanle Meng, Fu Zhang
Videos: bilibili   youtube   Code: Github   Star   Paper: Link

Trajectory generation for fully autonomous tail-sitter UAVs is challenging due to their nonlinear aerodynamics. This article presents the first fully autonomous tail-sitter UAV capable of high-speed navigation in unknown, cluttered environments. Enabled by LiDAR-based sensing and differential-flatness-based planning, it uses onboard computation for real-time control. We propose an optimization-based trajectory planning framework ensuring high-speed, collision-free, and dynamically feasible motion. To solve this nonlinear, constrained problem, we develop EFOPT, a feasibility-assured solver. Extensive simulations and real-world tests, including flights up to 15 m/s, validate its performance in diverse environments.

Authors: Guozheng Lu, Yunfan Ren, Fangcheng Zhu, Haotian Li, Ruize Xue, Yixi Cai, Ximin Lyu, Fu Zhang
Videos: bilibili   youtube   Paper: Link

In dynamic environments, robots require instantaneous detection of moving events with microseconds of latency. This task, known as moving event detection, is typically achieved using event cameras. While light detection and ranging (LiDAR) sensors are essential for robots due to their dense and accurate depth measurements, their use in event detection has not been thoroughly explored. Current approaches involve accumulating LiDAR points into frames and detecting object-level motions, resulting in a latency of tens to hundreds of milliseconds. We present a different approach called M-detector, which determines if a point is moving immediately after its arrival, resulting in a point-by-point detection with a latency of just several microseconds. M-detector is designed based on occlusion principles and can be used in different environments with various types of LiDAR sensors. Our experiments demonstrate the effectiveness of M-detector on various datasets and applications, showcasing its superior accuracy, computational efficiency, detection latency, and generalization ability.

Authors: Huajie Wu, Yihang Li, Wei Xu, Fanze Kong, Fu Zhang
Videos: bilibili   youtube   Code: Github   Star   Paper: Free access here

Swarm-LIO2 is a fully decentralized, plug-and-play, and bandwidth-efficient LiDAR-inertial odometry system for aerial swarms. It ensures accurate self and mutual state estimation, vital for tasks like cooperative exploration and search and rescue. Swarm-LIO2 uses a decentralized network to exchange minimal information, such as identity and mutual observations. It automatically detects new UAV teammates and initializes temporal offsets and global transformations. The system integrates LiDAR, IMU, and mutual observations in an efficient ESIKF framework, enhancing accuracy and consistency.

Authors: Fangcheng Zhu*, Yunfan Ren*, Longji Yin*, Fanze Kong, Qingbo Liu, Ruize Xue, Wenyi Liu, Yixi Cai, Guozheng Lu, Haotian Li, Fu Zhang
Videos: bilibili   youtube   Code: Github   Star   Paper: Link

R³LIVE++ is a LiDAR-inertial-visual fusion framework designed for robust and accurate state estimation and real-time radiance map reconstruction. It integrates real-time LiDAR-inertial odometry (LIO) for geometric structure and visual-inertial odometry (VIO) for radiance information. Building on R³ LIVE, it enhances localization and mapping accuracy by considering camera photometric calibration and online exposure time estimation. Extensive experiments on various datasets show R³ LIVE++'s superior accuracy and robustness compared to state-of-the-art SLAM systems. Applications based on the reconstructed maps include HDR imaging, virtual environment exploration, and 3D video gaming.

Authors: Jiarong Lin, Fu Zhang
Videos: bilibili   youtube   Code: Github   Star   Paper: Link

We present poweredflying ultra-underactuated LiDAR sensing aerial robot (PULSAR), an agile and selfrotating UAV whose three-dimensional position is fully controlled by actuating only one motor to obtain the required thrust and moment. The use of a single actuator effectively reduces the energy loss in powered flights. Consequently, PULSAR consumes 26.7% less power than the benchmarked quadrotor with the same total propeller disk area and avionic payloads while retaining a good level of agility. Augmented by an onboard LiDAR sensor, PULSAR can perform autonomous navigation in unknown environments and detect both static and dynamic obstacles in panoramic views without any external instruments.

Authors: Nan Chen, Fanze Kong, Wei Xu, Yixi Cai, Haotian Li, Dongjiao He, Youming Qin, Fu Zhang
Videos: bilibili   youtube   Code: Github   Star   Paper: Link

We tackle the theoretical and practical issues of trajectory generation and tracking control for tail-sitter UAVs. We focus on the differential flatness property using accurate UAV aerodynamic models, laying the groundwork for feasible trajectory generation and high-performance tracking. We found that tail-sitters are differentially flat with precise aerodynamic models across the entire flight envelope. This allows us to utilize high-fidelity models for accurate flight planning and control. We propose an optimization-based trajectory planner that accounts for kinodynamic constraints, singularity-free conditions, and actuator saturation. To track the trajectory, we developed a global, singularity-free, minimally parameterized on-manifold MPC. Our algorithms were tested on the quadrotor tail-sitter prototype "Hong Hu" and demonstrated effectiveness in various real-world experiments, including agile SE(3) flight, typical maneuvers, and aggressive aerobatics.

Authors: Guozheng Lu, Yixi Cai, Nan Chen, Fanze Kong, Yunfan Ren, Fu Zhang
Videos: bilibili   youtube   Paper: Link

BALM2 is an efficient and consistent bundle adjustment method for LiDAR sensors, addressing the simultaneous determination of sensor poses and scene geometry. It uses edge and plane features to represent geometry, minimizing the Euclidean distance from raw points to these features. This formulation allows for analytical solutions, reducing optimization dimensions. The method introduces point clusters to encode raw points with compact parameters, enhancing optimization efficiency. A second-order BA solver is developed, providing consistent pose estimates and uncertainty evaluation. This solver avoids raw point enumeration in optimization steps.

Authors: Zheng Liu, Xiyuan Liu, Fu Zhang
Videos: bilibili   youtube   Code: Github   Star   Paper: Link

We present D-Map, an efficient occupancy mapping framework for high-resolution LiDAR sensors. To address computational efficiency challenges, I propose three main innovations. First, I use depth images to determine occupancy states instead of traditional ray-casting. Second, an efficient on-tree update strategy is introduced, avoiding redundant visits to small cells and reducing updates. Third, I remove known cells during each update, leveraging LiDAR's low false alarm rate, enhancing efficiency and introducing a decremental property. Theoretical analysis and extensive benchmark experiments demonstrate D-Map's superior efficiency and comparable accuracy. Real-world applications include real-time occupancy mapping on a handheld device and an aerial platform.

Authors: Yixi Cai, Fanze Kong, Yunfan Ren, Fangcheng Zhu, Jiarong Lin, Fu Zhang
Videos: bilibili   youtube   Code: Github   Star   Paper: Link

In recent years, advancements in LiDAR technology have made 3D LiDAR sensors more accessible, spurring interest in multi-sensor fusion for SLAM research. However, datasets with aerial downward-looking views are scarce. To address this, I introduce the MARS-LVIG dataset, providing unique aerial LiDAR-Visual-Inertial-GNSS data from altitudes between 80 m and 130 m. The dataset includes 21 sequences across diverse environments, with UAV speeds from 3 m/s to 12 m/s, covering areas up to 577,000 m². It features a synchronized sensor package and includes ground truth for localization and mapping. Download at: https://mars.hku.hk/dataset.html.

Authors: Haotian Li, Yuying Zou, Nan Chen, Jiarong Lin, Xiyuan Liu, Wei Xu, Chunran Zheng, Rundong Li, Dongjiao He, Fanze Kong, Yixi Cai, Zheng Liu, Shunbo Zhou, Kaiwen Xue, Fu Zhang
Dataset: dataset   Paper: Link

Accurate and robust place recognition is crucial for robot navigation, yet achieving full pose invariance across diverse scenes is challenging. In this work, I propose the Binary Triangle Combined (BTC) descriptor, a novel global and local combined descriptor. We extract keypoints from a point cloud and form unique triangles from any three keypoints. The triangle sides form a global descriptor, while a binary descriptor captures local point distributions. This combination allows for natural similarity determination and accurate relative pose estimation. Our BTC descriptor enhances accuracy, efficiency, and robustness by combining global and local geometric information. Extensive comparisons with state-of-the-art methods on diverse datasets demonstrate BTC's superior adaptability and precision, especially in challenging scenarios with significant viewpoint variations.

Authors: Chongjian Yuan, Jiarong Lin, Zheng Liu, Hairuo Wei, Xiaoping Hong, Fu Zhang
Videos: youtube   Code: Github   Star   Paper: Link

In this article, I propose ImMesh, a novel LiDAR(-inertial) odometry and mapping framework for real-time simultaneous localization and meshing. ImMesh consists of four tightly-coupled modules: receiver, localization, meshing, and broadcaster. The localization module processes sensor data to estimate sensor pose and dynamically update the map. The meshing module incrementally reconstructs triangle meshes from registered LiDAR scans using an efficient voxel structure. This voxel-wise meshing operation is designed for efficiency, projecting 3D points to 2D planes and reconstructing triangles incrementally. ImMesh publishes real-time odometry, maps, and meshes via the broadcaster. This framework uniquely achieves large-scale, real-time mesh reconstruction using only a standard CPU.

Authors: Jiarong Lin, Chongjian Yuan, Yixi Cai, Haotian Li, Yunfan Ren, Yuying Zou, Xiaoping Hong, Fu Zhang
Videos: bilibili   youtube   Code: Github   Star   Paper: Link

Point-LIO is a robust and high-bandwidth LiDAR-inertial odometry with the capability to estimate extremely aggressive robotic motions. In experiments, it is able to provide accurate, high-frequency odometry (4-8 kHz) and reliable mapping under severe vibrations and aggressive motions with high angular velocity beyond the IMU measuring ranges. And Point-LIO is computationally efficient, robust, versatile on public datasets with general motions. As an odometry, Point-LIO could be used in various autonomous tasks, such as trajectory planning, control, and perception, especially in cases involving very fast ego-motions or requiring high-rate odometry output and mapping.

Authors: Dongjiao He, Wei Xu, Nan Chen, Fanze Kong, Chongjian Yuan, Fu Zhang
Videos: bilibili   youtube   Code: Github   Star   Paper: Link

FAST-LIO2 is computationally-efficient (e.g., up to 100 Hz odometry and mapping in large outdoor environments), robust (e.g., reliable pose estimation in cluttered indoor environments with rotation up to 1000 deg/s), versatile (i.e., applicable to both multi-line spinning and solid-state LiDARs, UAV and handheld platforms, and Intel and ARM-based processors), while still achieving higher or comparable accuracy with existing methods.

Authors: Wei Xu, Yixi Cai, Dongjiao He, Jiarong Lin, Fu Zhang
Videos: bilibili   youtube1 youtube2   Code: Github   Star   Paper: Link