TL;DR
This paper presents a novel software and hardware system designed to facilitate hands-on quadcopter control education for undergraduate students. By enabling simultaneous, independent operation of multiple quadcopters within a managed classroom environment, the system bridges the gap between theoretical learning and practical application, fostering deeper understanding and engagement.
What Problem Does This Paper Address?
Teaching control engineering effectively often involves hands-on experimentation. However, traditional approaches face challenges: instructor-led experiments limit student interaction, while fully independent "hands-on" setups can be difficult to manage for larger groups. Quadcopters offer an ideal platform for control education due to their relevance and the complexity of their control systems. The problem this paper addresses is the lack of a robust, scalable, and easy-to-manage system that allows a moderate number of undergraduate students to simultaneously gain practical experience controlling quadcopters in a classroom setting. Existing systems are often research-focused, not designed for educational throughput, or lack the necessary management layers for a smooth classroom experience.
Methods
The authors developed a multi-agent system built around the open-source Crazyflie 2.0 quadcopter. The core of their solution lies in a distributed software architecture leveraging Robot Operating System (ROS). Key components include:
- Hardware Modifications: 3D-printed parts were added to the Crazyflie 2.0 for attaching retro-reflective markers for motion capture and for better battery placement.
- Motion Capture (Mocap) System: A commercial Mocap system (Vicon, Qualisys, OptiTrack) is used for high-frequency, accurate localization of multiple quadcopters in a shared coordinate system.
- ROS Architecture: The system is built using ROS nodes, allowing for distributed operation across multiple computers. This enables students to work independently on their own quadcopters.
- System Config GUI: A teacher-facing GUI allows for easy allocation of flying zones, quadcopters, and student computers, simplifying classroom management.
- Flying Agent Client: This node manages individual quadcopters, handling Mocap data, controller selection, safety mechanisms, and communication with the quadcopter.
- Default and Custom Controllers: Students implement their control algorithms in "Custom Controller" nodes, while a "Default Controller" acts as a fallback for safety.
- Cascaded Control Architecture: The control system follows a common cascaded approach, with an inner loop for body rates (on-board) and an outer loop for position and heading (off-board, implemented by students).
- Qt-based GUIs: User-friendly interfaces are provided for both teachers (System Config GUI) and students (Flying Agent GUI) to interact with the system.
Datasets
Not discussed in the paper. The system is designed for real-time experimentation rather than training on pre-existing datasets.
Key Insights and Ideas
- Bridging Theory and Practice: The system effectively connects theoretical control concepts learned in lectures to a tangible, real-world system, significantly enhancing student understanding and motivation.
- Productive Failure: The system is designed to allow for "productive failure," where students can observe the consequences of their control design choices (e.g., instability, crashes) in a safe and managed way, leading to deeper learning.
- Scalability and Manageability: The ROS-based distributed architecture and the System Config GUI are crucial for managing a classroom of multiple students and quadcopters simultaneously, a significant improvement over previous approaches.
- Collaborative Learning: The setup encourages collaboration among students, fostering soft skills alongside technical expertise.
- Flexibility: The open-source nature and modular ROS design allow for easy adaptation and extension for teaching advanced control concepts or for research projects.
Key Results
The paper describes a successful implementation of this system in a hands-on control course for 32 undergraduate electrical engineering students at ETH Zurich. Survey results indicated that 90% of students found the course useful for understanding control systems theory. Student feedback highlighted the ease with which they could design controllers for a complex system using basic control theory tools and the value of the hands-on experience. The system demonstrated its ability to manage multiple quadcopters safely and effectively within a classroom environment.
Limitations and Future Work
- Limitations: While the system incorporates safety mechanisms, the authors acknowledge that crashes can still occur, though the low inertia of the Crazyflie mitigates damage. The accuracy of the Mocap system is crucial, and its setup can be time-consuming.
- Future Work: The authors suggest that the platform can be used for teaching advanced control concepts such as Robust Control, Model Predictive Control, or Distributed Control. It also serves as a testbed for Bachelor's and Master's projects, combining theoretical, practical, and research considerations. The system is also being used for validation of learning-based control. The modular design of the Qt GUIs also allows for future expansion and customization.
Summary updated