An autonomous swarm of micro flying robots with range-based relative localization.

Shushuai Li, Mario Coppola, Christophe De Wagter, Guido C. H. E. de Croon

Abstract

Researchers have developed a fully autonomous, onboard relative localization system that allows swarms of tiny, 33-gram drones to navigate without any external infrastructure like GPS or motion-capture cameras. By combining Ultra-Wideband (UWB) ranging with a clever "self-regulating" estimation theory, they achieved high-precision flight for up to 13 drones simultaneously on extremely limited hardware.

Introduction: The Challenge of Micro-Drone Swarms

For a swarm of drones to fly together without crashing, every drone needs to know where its neighbors are—a task called relative localization. While large drones can carry heavy sensors like LiDAR or powerful cameras, "micro" drones (like the Bitcraze Crazyflie) are severely limited by their weight (33g) and processing power (a modest 168MHz processor with only 192KB of RAM).

Existing solutions often fail in these resource-constrained scenarios. Infrastructure-based systems like GPS are unavailable indoors, and vision-based systems are often too computationally expensive or sensitive to lighting. Furthermore, using radio ranging (UWB) to find positions is mathematically complex; there are "unobservable" states where a drone lacks sufficient geometric information to determine its exact relative position.

System Architecture and Methods

The proposed system addresses these challenges through three integrated components:

1. Scalable Ranging Protocol

Instead of traditional one-to-one communication, which scales poorly as swarm size increases, the authors developed a broadcast-based UWB protocol. Each drone periodically broadcasts a "ranging message" containing its own state (velocity, yaw, height) and timestamps. This allows the entire swarm to update distances concurrently at 16Hz.

2. Onboard Data Fusion

The researchers implemented a highly efficient Extended Kalman Filter (EKF) that runs entirely on the drone's microcontroller. This filter fuses UWB distance measurements with shared state data from neighbors and local sensor data (IMU, optical flow, and laser altimeter).

System architecture showing the onboard sensors and the data flow for relative estimation.

3. Stability and "Self-Regulated" Observability

A key theoretical contribution is the analysis of system "unobservability" using Lie derivatives. The researchers discovered a "self-regulated" property: if a drone's position estimate begins to drift due to unobservability (e.g., during a static hover), the resulting control actions naturally move the drone in a way that restores observability, correcting the error.

Relative coordinate frames between two drones (i and j) used for relative state estimation.

Experimental Validation

The system was evaluated through both numerical simulations and real-world flight tests:

  • Simulations: 50 random tests verified EKF convergence and accuracy under various noise conditions.
  • Flight Tests: Swarms of 3, 5, and 13 Crazyflie 2.1 drones were tested in formation flight.
  • Ground Truth: An OptiTrack motion capture system provided a reference for measuring accuracy during experiments.

Key Results

  • Precision: Achieved a position error of less than 0.2 meters in real-world flights.
  • Scalability: Successfully demonstrated a swarm of 13 drones flying in formation—the first time such a large micro-swarm has been localized purely via onboard ranging.
  • Convergence: The system converges from unknown initial positions to an accurate estimate in approximately 20 seconds.
  • Efficiency: The UWB ranging technique achieved a mean absolute error of just 0.03 meters, outperforming state-of-the-art methods while requiring fewer resources.

Discussion and Future Work

The current system is cooperative, relying on neighbors to share velocity and yaw data. Consequently, it cannot localize non-cooperative drones. While "self-regulation" handles most scenarios, extreme cases like perfectly static, symmetric formations might still experience yaw drift. Future research could integrate minimal visual cues or decentralized control laws to mitigate these edge cases. The open-sourcing of this code provides a robust foundation for future search-and-rescue and exploration micro-swarms.

Summary updated

Abstract

Accurate relative localization is an important requirement for a swarm of robots, especially when performing a cooperative task. This paper presents an autonomous multi-robot relative positioning technique. An Extended Kalman filter (EKF) uses onboard sensing of velocity, yaw rate, and height as inputs, and then estimates the relative position of other robots by fusing these quantities with ranging measurements obtained from onboard ultra wide-band (UWB). Specifically, innovations involve fast-ranging communication (333Hz for 2 robots), an automatic initialization procedure, proofs and demonstrations of consistent estimation convergence under control commands such as formation flight. Simulations concisely show the high precision, efficiency, and stability of the proposed localization method. Real-world experiments are conducted on a team of 5 Crazyflie2 quadrotors, demonstrating autonomous formation flight and coordinated flight through a window. All results indicate the effectiveness of the proposed relative positioning method for multi-robot systems. Video and code can be found at \textnormal{\url{https://shushuai3.github.io/autonomous-swarm/}}
BibTeX
@article{li2020autonomous,
  author = {Shushuai Li and Mario Coppola and Christophe De Wagter and Guido C. H. E. de Croon},
  title = {{An autonomous swarm of micro flying robots with range-based relative localization.}},
  journal = {IEEE Robotics Autom. Lett.},
  year = {2020},
}
Figure