Preserving Relative Localization of FoV-Limited Drone Swarm via Active Mutual Observation

Lianjie Guo, Zaitian Gongye, Ziyi Xu, Yingjian Wang, Xin Zhou, Jinni Zhou, Fei Gao

Abstract

Relative state estimation is crucial for vision-based swarms to estimate and compensate for the unavoidable drift of visual odometry. For autonomous drones equipped with the most compact sensor setting—a stereo camera that provides a limited field of view (FoV)—the demand for mutual observation for relative state estimation conflicts with the demand for environment observation. To balance these two demands for FoV-limited swarms by acquiring mutual observations with a safety guarantee, this paper proposes an active localization correction system, which plans camera orientations via a yaw planner during the flight. The yaw planner manages the contradiction by calculating suitable timing and yaw angle commands based on the evaluation of localization uncertainty estimated by the Kalman Filter. Simulation validates the scalability of our algorithm. In real-world experiments, we reduce positioning drift by up to 65% and managed to maintain a given formation in both indoor and outdoor GPS-denied flight, from which the accuracy, efficiency, and robustness of the proposed system are verified.

Introduction

In the realm of autonomous drone swarms, maintaining precise relative positioning is a prerequisite for formation flight and collision avoidance. Most micro-drones rely on Visual-Inertial Odometry (VIO) for localization; however, VIO is prone to cumulative drift over time. While mutual observation—drones detecting one another—can correct this drift, it presents a significant challenge for drones with limited Field-of-View (FoV) sensors, such as forward-facing stereo cameras.

A fundamental conflict of interest arises: to fly safely, a drone must look forward at its trajectory to avoid obstacles; to correct its location, it often needs to look sideways or backward at its teammates. This paper introduces an active localization correction system that intelligently manages this trade-off through strategic yaw planning, allowing drones to "look" at each other only when necessary without compromising flight safety.

System Architecture

The proposed system utilizes a centralized architecture where a leader drone manages the swarm's collective state through three primary components:

  1. State Estimation via Kalman Filter: A centralized Kalman Filter (KF) tracks the position drifts of every drone in the swarm. The filter's covariance matrix serves as a mathematical representation of the uncertainty between any two drones.
  2. Active Yaw Planner: Instead of constant observation, the system treats mutual observation as a resource. A planner monitors the trace of the covariance matrix. When uncertainty exceeds a specific threshold, the planner selects a "target" and an "observer" drone, calculating a specific yaw command to facilitate a correction.
  3. Safety and Visibility Constraints: Before a drone is allowed to turn away from its flight path, the system performs a safety check. It calculates if the drone has enough "known" space ahead to safely maneuver during the time its camera is diverted. It also checks for environmental occlusions, such as trees, that might block the view.

Key Methodologies

The Confidence Ellipse

Because the target drone's position is also drifting, the observer cannot simply point at a single coordinate. The planner calculates a "confidence area"—an elliptical region based on a chi-square distribution (95% probability) where the target is statistically likely to be. The yaw planner ensures the observer's FoV sweeps across this entire area to guarantee detection.

Implicit Correction

A significant property of the Kalman Filter used in this framework is implicit correction. Observing one drone can reduce the uncertainty of other drones in the swarm, provided they have had mutual observations in the past. This creates a network effect that improves the localization of the entire swarm even with sparse observations.

Experimental Validation

Simulation and Scalability

The researchers validated the scalability of the Kalman Filter and yaw planner with simulations of up to 100 drones. The results showed high efficiency: for a 100-drone swarm, the KF update takes approximately 5ms and the yaw planning roughly 2ms, making the system viable for real-time deployment.

Indoor and Outdoor Testing

  • Indoor Experiments: Using a motion capture system for ground truth, the system was tested with a four-drone swarm. The active localization system reduced total relative positioning drift by up to 65% compared to standard VIO.
  • Outdoor Field Tests: The system was deployed in a forest environment where three drones flew in a line formation without GPS. The drones successfully deformed their formation to avoid obstacles while using active yawing to maintain relative position accuracy sufficient for stable formation flight.

Conclusion and Future Work

This research demonstrates that active mutual observation can significantly mitigate VIO drift in FoV-limited drone swarms. By treating localization as an active planning problem rather than a passive sensing task, the system achieves high accuracy with minimal sensor hardware.

Future work will address the limitations of the current centralized architecture to prevent single points of failure and explore path planning—adjusting the actual flight trajectory rather than just the yaw—to further facilitate mutual observations in complex environments. Additionally, the impact of communication latency in high-density swarms remains a key area for further investigation.

Summary updated

Abstract

Relative state estimation is crucial for vision-based swarms to estimate and compensate for the unavoidable drift of visual odometry. For autonomous drones equipped with the most compact sensor setting -- a stereo camera that provides a limited field of view (FoV), the demand for mutual observation for relative state estimation conflicts with the demand for environment observation. To balance the two demands for FoV limited swarms by acquiring mutual observations with a safety guarantee, this paper proposes an active localization correction system, which plans camera orientations via a yaw planner during the flight. The yaw planner manages the contradiction by calculating suitable timing and yaw angle commands based on the evaluation of localization uncertainty estimated by the Kalman Filter. Simulation validates the scalability of our algorithm. In real-world experiments, we reduce positioning drift by up to 65% and managed to maintain a given formation in both indoor and outdoor GPS-denied flight, from which the accuracy, efficiency, and robustness of the proposed system are verified.
BibTeX
@inproceedings{guo2024preserving,
  author = {Lianjie Guo and Zaitian Gongye and Ziyi Xu and Yingjian Wang and Xin Zhou and Jinni Zhou and Fei Gao},
  title = {{Preserving Relative Localization of FoV-Limited Drone Swarm via Active Mutual Observation}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2024},
  url = {https://openreview.net/forum?id=lnjCQ8prYo},
}
Figure