FACT: Fast and Active Coordinate Initialization for Vision-based Drone Swarms

TL;DR

This paper addresses the challenge of fast, robust initial relative pose estimation for SWaP-constrained vision-based drone swarms without external localization. It decomposes the problem into rotation estimation using a dual semidefinite relaxation (SE-Sync) on $Z\in Sym(2N)$ to obtain a global rotation in $SO(2)^N$, followed by translation estimation via the Hungarian algorithm on anonymous observations. An active planning module ensures multiple non-colliding observations, and a DG-VDT pipeline provides drone detections using stereo cameras and IMUs with onboard computation. Experiments on simulated and real-world drone swarms demonstrate real-time performance, robustness to noise, and improved global optimality compared to local optimization methods, with open-source code released for reproducibility.

Abstract

Swarm robots have sparked remarkable developments across a range of fields. While it is necessary for various applications in swarm robots, a fast and robust coordinate initialization in vision-based drone swarms remains elusive. To this end, our paper proposes a complete system to recover a swarm's initial relative pose on platforms with size, weight, and power (SWaP) constraints. To overcome limited coverage of field-of-view (FoV), the drones rotate in place to obtain observations. To tackle the anonymous measurements, we formulate a non-convex rotation estimation problem and transform it into a semi-definite programming (SDP) problem, which can steadily obtain global optimal values. Then we utilize the Hungarian algorithm to recover relative translation and correspondences between observations and drone identities. To safely acquire complete observations, we actively search for positions and generate feasible trajectories to avoid collisions. To validate the practicability of our system, we conduct experiments on a vision-based drone swarm with only stereo cameras and inertial measurement units (IMUs) as sensors. The results demonstrate that the system can robustly get accurate relative poses in real time with limited onboard computation resources. The source code is released.

Figures (9)

• Figure 1: Schematic of vision-based coordinate initialization. Observations were sketched with bi-directional arrows. We utilize the dual semi-definite relaxation to formulate rotation estimation as a semi-definite programming (SDP) problem. Hungarian algorithm is used to solve the correspondence and come up with the translation estimations.
• Figure 2: Challenges posed by anonymous, vision-based measurements.
• Figure 3: System overview. The main modules that our work focuses on are marked red and encapsulated by dashed lines.
• Figure 4: The implementation of the DG-VDT module.
• Figure 5: Schematic for a pair of mutual observations.