TL;DR
This paper presents a novel Model Predictive Control (MPC) approach for controlling a formation of airships to cooperatively track a moving subject. It addresses the unique non-holonomic constraints of airships, which differ significantly from multi-rotors, enabling safer and quieter aerial motion capture in natural environments.
What Problem Does This Paper Address?
Aerial motion capture (MoCap) is crucial for tasks like tracking animals in their natural habitats. While multi-rotor drones have been used, they are noisy, have limited flight times, and can be unsafe around wildlife. Airships offer a quieter, safer, and longer-endurance alternative. However, airships have complex, coupled dynamics (e.g., orientation and flight direction are linked, and they are affected by wind) that make existing multi-rotor formation control strategies inapplicable. The core challenge is to develop a formation control system for airships that can maintain optimal MoCap conditions – keeping the subject centered in each camera, at a safe distance, and with specific angular separation between airships – despite these unique constraints.
Methods
The authors tackle this by first deriving analytical solutions for airship formation control in simplified 2D scenarios, exploiting a periodic relationship between airship airspeed and its distance to the subject. This forms the basis for a more realistic 3D model that accounts for airship aerodynamics, including lateral angle of attack, roll, and pitch. They then formulate an optimization problem to find optimal trajectories and develop a Model Predictive Control (MPC) framework. This MPC controller is designed to satisfy the MoCap requirements while respecting the airship's physical constraints. The approach uses a numerical solver to find optimal control inputs over a finite horizon, allowing for real-time adjustments.
Datasets
Not discussed in the paper. The authors mention using simulated data for experiments and a real-world flight with a single airship.
Key Insights and Ideas
- Airships are a viable alternative for wildlife MoCap: Their quiet operation, safety, and endurance make them superior to multi-rotors for natural environments.
- Airship dynamics are fundamentally different: The coupled nature of airship orientation and flight direction, along with wind effects, necessitates a new control paradigm beyond multi-rotor approaches.
- Exploiting airspeed-distance periodicity: A key insight is the periodic relationship between an airship's airspeed and its distance to the subject, which allows for analytical solutions to maintain optimal camera views.
- MPC for robust formation control: MPC proves effective in handling the complex dynamics and constraints, allowing for dynamic adjustments to maintain formation and tracking goals.
- 3D effects matter: While 2D analysis provides a foundation, accounting for realistic 3D physics like pitch and roll significantly impacts achievable orbits and requires a numerical approach.
Key Results
- Simulation experiments: The MPC controller demonstrated effective formation control across various scenarios, including different formation sizes (up to 6 airships), wind conditions (up to 3.0 m/s), and subject motion (including sharp turns).
- Tracking accuracy: For formations of 2 or more airships, tracking accuracy and uncertainty were significantly reduced compared to a single airship.
- Subject visibility: The system maintained subject visibility in camera frames for over 90% of the time, even under challenging wind conditions.
- Real-world demonstration: A real-world experiment with a single airship validated the MPC's performance in maintaining camera pointing under realistic wind conditions (1-2 m/s with gusts up to 4 m/s), showing good agreement with simulations.
- Open-source code and video: The authors have released their code and a video demonstrating the method.
Limitations and Future Work
- Analytical boundary for 3D: While analytical solutions were derived for 2D, the exact boundary for subject velocity in the 3D case, considering pitch angle, is not analytically determined and depends on various parameters.
- Collision and obstacle avoidance: The current work focuses on formation control and subject tracking. Future work will investigate collision and obstacle avoidance within the airship formation context.
- Generalization to other vehicles: The authors plan to generalize their approach to other non-holonomic vehicles, such as fixed-wing aircraft.
- Subject velocity estimation: The current MPC assumes a known or estimated subject velocity. More robust estimation techniques could be explored.
Summary updated