Swarm-LIO2: Decentralized, Efficient LiDAR-inertial Odometry for UAV Swarms

TL;DR

Swarm-LIO2 addresses the challenge of scalable, robust state estimation for aerial UAV swarms by delivering a fully decentralized LiDAR-inertial odometry framework that supports plug-and-play teammates. It introduces three key innovations: reflectivity-based LiDAR teammate detection with trajectory matching, a decentralized factor-graph calibration for not-directly-observed teammates, and a marginalization-based ESIKF state estimator that fuses LiDAR, IMU, and mutual observations while compensating temporal delays. The approach achieves online joining/leaving, low-bandwidth inter-UAV communication, and scalable computation, with demonstrated stability in simulations up to 40 UAVs and real-world experiments with five UAVs, including degenerate and cluttered environments. Collectively, Swarm-LIO2 provides a practical infrastructure for cooperative exploration, target tracking, collision avoidance, and payload transportation in UAV swarms, with potential extensions to loop closures for long-term drift reduction.

Abstract

Aerial swarm systems possess immense potential in various aspects, such as cooperative exploration, target tracking, search and rescue. Efficient, accurate self and mutual state estimation are the critical preconditions for completing these swarm tasks, which remain challenging research topics. This paper proposes Swarm-LIO2: a fully decentralized, plug-and-play, computationally efficient, and bandwidth-efficient LiDAR-inertial odometry for aerial swarm systems. Swarm-LIO2 uses a decentralized, plug-and-play network as the communication infrastructure. Only bandwidth-efficient and low-dimensional information is exchanged, including identity, ego-state, mutual observation measurements, and global extrinsic transformations. To support the plug-and-play of new teammate participants, Swarm-LIO2 detects potential teammate UAVs and initializes the temporal offset and global extrinsic transformation all automatically. To enhance the initialization efficiency, novel reflectivity-based UAV detection, trajectory matching, and factor graph optimization methods are proposed. For state estimation, Swarm-LIO2 fuses LiDAR, IMU, and mutual observation measurements within an efficient ESIKF framework, with careful compensation of temporal delay and modeling of measurements to enhance the accuracy and consistency.

Figures (21)

• Figure 1: An aerial swarm system constituted of five UAVs flying in the wild in which Swarm-LIO2 serves as the self and mutual state estimator. More details can be found in the attached video at https://youtu.be/Q7cJ9iRhlrY
• Figure 2: The illustration of the swarm state estimation problem.
• Figure 3: Framework of the proposed state estimation system for aerial swarm systems. The dashed arrow lines mean the messages are sent only once, while the solid arrow lines mean the messages are sent constantly at the scan rate.
• Figure 4: Illustration of the initialization for newly detected objects, point-cloud is colored by reflectivity. Here the self-UAV needs to detect and identify other teammate UAVs in its FoV. The sphere represents the predicted region and the center of the bounding box represents the updated position of the tracker. (a) Reflectivity filtering. (b) Outlier rejection by discarding objects with too large size. (c) Track real potential teammates and accumulate the trajectory. (d) After trajectory matching, the object is identified as a teammate UAV with a correct UAV ID.
• Figure 5: The illustration of the decentralized factor graph optimization-based global extrinsic calibration. To handle the gauge freedom of the factor graph, we insert a prior factor as $G_i = I$. Note that the factors $^{G_i}\breve{\mathbf T}_{G_j}$ are global extrinsic transformations received from teammate UAVs or obtained by direct trajectory matching on the self-UAV.