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

Yuan Li,Anke Zhao,Yingjian Wang,Ziyi Xu,Xin Zhou,Jinni Zhou,Chao Xu,Fei Gao

2024-03-20

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.

Robotics

What problem does this paper attempt to address?

The paper primarily focuses on addressing the problem of coordinate initialization for visual drone swarms under Size, Weight, and Power (SWaP) constraints, specifically how to quickly and robustly determine the relative positions and orientations among drones in the swarm. Specifically, the main challenges faced by the research include:

1. **Limited Field-of-View (FoV)**: The sensor's field of view of the drones is limited, which restricts their ability to observe each other.

2. **Anonymous Measurements**: The observed information cannot be directly associated with the identity of specific drones.

3. **Safety**: During the initialization process, it is necessary to avoid collisions with other drones or obstacles.

To overcome these challenges, the paper proposes a systematic solution that includes three main components:

- **Observation Module**: Drones increase their observation range by rotating in place to compensate for the limitations of the field of view angle.

- **Pose Estimation**: The non-convex rotation estimation problem is transformed into a Semi-Definite Programming (SDP) problem, and the Hungarian algorithm is used to solve the relative translation and the correspondence between observations and drone identities.

- **Active Planning Module**: Selects target positions and generates feasible trajectories to ensure that drones can perform the next observation safely.

The system has been experimentally validated on a visual drone swarm equipped with stereo cameras and Inertial Measurement Units (IMUs) as sensors. The results show that the system can obtain accurate relative pose information in real-time and robustly, even with limited onboard computational resources. Additionally, compared to local optimization methods, this approach can stably achieve a globally optimal solution and significantly reduce the solving time.