As the core field of unmanned system research, drone drive and control technology is undergoing a profound transformation from single machine autonomy to cluster collaboration. Whether it is reconnaissance and strike in the military field, logistics and distribution in the civilian field, or agricultural crop protection, the performance boundary and application scenarios of drones directly depend on the advanced level of the driving and control scheme.
A quadcopter drone is a structurally simple and high-performance vertical takeoff and landing aircraft, but it is also an underactuated system with four inputs and six outputs, and a multivariable, strongly coupled, statically unstable nonlinear system. These characteristics pose significant challenges to its control technology.
Early drone control mainly used PID (Proportional Integral Derivative) controllers, which decomposed the six degree of freedom system into two linear subsystems through feedback linearization: the inner loop attitude subsystem and the outer loop position subsystem, connected by a nonlinear coupling term.
The PID control method is simple and reliable, but it seems inadequate when dealing with nonlinear and strongly coupled dynamics.
With the increasing complexity of tasks, more and more advanced control algorithms are being introduced into the field of drone driving and control:
- Fuzzy PID control: Combining fuzzy logic with PID control improves the system's adaptability to nonlinear and time-varying uncertainties.
- Particle swarm fuzzy PID control: By using particle swarm optimization algorithm to optimize the fuzzy PID parameters, the response speed and anti-interference ability of the control system are further improved.
- Deep reinforcement learning control: In recent years, the combination of deep reinforcement learning and quadcopter drone control has shown great potential. For example, combining deep reinforcement learning with bidirectional thrust control of quadcopter drones can achieve fast hovering in intense motion, smoother execution of actions, and smaller state fluctuations.
- Model Predictive Control (MPC): For the height and attitude control problems of the transition mode of tiltrotor unmanned aerial vehicles, the MPC method iteratively optimizes to modify the controller and its weight, significantly improving the dynamic characteristics of the transition mode.
The electronic speed control module of the drone is the core of the drone power system, responsible for controlling the speed and rotation of the brushless DC motor. ESC includes power level, current sensing circuit, microcontroller, and communication interface with flight control system.
The most commonly used motor control technology by ESC manufacturers currently is Field Oriented Control (FOC), which can handle rapid acceleration changes without generating instability, allowing drones to perform complex maneuvers while maximizing efficiency.
With the complex changes in tasks and environments, a single drone is no longer sufficient to meet mission requirements. Drone swarms composed of multiple drones have gradually gained attention from countries around the world due to their excellent performance such as high efficiency, robustness, multifunctionality, and scalability.
The collaborative control of drone clusters has formed three mainstream control structures:
- Centralized control: It is necessary to designate a drone as the core of the cluster to process all information and issue control instructions to all drones. This method is easy to implement and has high formation accuracy, but it has poor robustness and high communication load.
- Distributed control: Without an information processing core, control of the cluster is achieved solely through information exchange with neighboring nodes. This method significantly reduces the amount of information exchange and computation in the cluster, making it more stable and flexible.
- Distributed control: adopting a control method that corresponds one-to-one between the controller and the drone, without a control center, and there is no communication relationship between each node. The implementation is simple and scalable, but the adaptability and robustness are poor.
- Navigation and Follow Method: Designate a drone in the cluster as the leader, and other drones follow its trajectory to maintain a certain relative distance in flight. This method is simple to implement and has high formation accuracy, but the system robustness is poor.
- Virtual structure method: Treat the entire cluster as a whole, perform kinematic and dynamic analysis on the virtual structure, and control the drone to track the corresponding target points. This method has high control accuracy and a certain degree of fault tolerance, but it limits the flexibility of the system.
- Behavioral approach: Set basic behaviors such as formation, obstacle avoidance, and following for each drone in the cluster, and adopt different cluster behaviors based on different external information. This method is easy to implement and flexible, but the overall behavior is difficult to clarify.
- Consistency method: Drones gradually achieve consistency in their motion state variables through information sharing between adjacent nodes. This method is suitable for large-scale clusters, but the algorithm design is complex.
Academician Wang Yaonan from Hunan University and Professor Fang Yongchun from Nankai University have collaborated to propose a performance function guided deep reinforcement learning control method for unmanned aerial vehicle cluster systems. The method evaluates the demonstration experience of performance functions and explores learning strategies to ensure efficient and reliable strategy updates.
Professor Duan Haibin from Beihang University and Academician Li Ming from Shenyang Aircraft Design and Research Institute of China Aviation Industry Corporation have collaborated to propose a phase change control method for unmanned aerial vehicle (UAV) swarms based on a bird swarm self-propelled particle model. The motion model of the bird swarm is constructed by designing a velocity holding term and a potential energy gradient term.
In response to the problems of low training efficiency and slow convergence speed in dynamic route planning for low altitude aircraft, researchers propose a deep Q-network dynamic route planning algorithm based on target oriented course learning and priority experience replay strategy. By introducing a course learning mechanism and setting a target guided maneuver strategy, the algorithm improves the training speed of the algorithm while optimizing the floatability of the planned route.
Innovative driving solutions are constantly emerging, such as a multi rotor control system that uses horizontal thrust to control the high-speed flight of the aircraft. Based on the multi rotor unmanned aerial vehicle, a thruster based horizontal thrust mechanism is installed at each rotor, and the flight control system cooperates to control the rotor speed and thruster horizontal output, so that the rotor and thruster work together to generate a joint force.
This design can significantly reduce the pitch angle of the aircraft during flight, reduce energy loss, and improve response speed, control accuracy, and stability.
| Application Field | Technical Requirements / Applicable Drive Control Schemes |
|---|---|
| Logistics distribution | High precision hovering, obstacle avoidance, bidirectional thrust control for path planning, and dynamic route planning algorithm |
| Stable flight during power inspection, anti-interference ability | Particle swarm fuzzy PID control, sensor fusion algorithm |
| Agricultural plant protection cluster collaboration, full coverage path | Distributed cluster control, behavioral law formation |
| Adaptability to complex emergency rescue environments, robustness for amphibious drive control in land and air | Reinforcement learning adaptive control |
| Geographic mapping high-precision positioning, route tracking | Consistency formation, preset performance control |
Drone control technology still faces many challenges. In the field of low altitude aircraft, the core challenge is how to support high-capacity flight characterized by "heterogeneity, high density, high frequency, and high complexity", and achieve rapid development of low altitude economy scale, sustainability, and high quality under the premise of safety and controllability.
The key problem that urgently needs to be solved in cluster collaborative control technology is to reduce the communication load of centralized control and enhance the robustness of cluster control; Optimize the algorithm design of distributed control and improve the accuracy of cluster control; Enhance the control effect of decentralized control and improve cluster stability.
The future development of drone drive and control technology will focus on the following directions:
- Deep integration of artificial intelligence: AI technologies such as deep learning and reinforcement learning are deeply integrated with driving and control algorithms to improve system autonomy and environmental adaptability.
- Heterogeneous system collaboration: Unmanned aerial vehicles, unmanned ships, and other heterogeneous platforms collaborate and control to achieve more complex cross domain combat and operational capabilities.
- Anti interference and safety: Enhance the anti-interference ability in complex electromagnetic environments to ensure system safety and reliability.
- Breakthrough in biomimetic intelligence: Imitate the collective behavior mechanism of biological groups such as birds and insects, and develop more efficient and flexible cluster driving and control algorithms.
The development of drone drive and control technology is an interdisciplinary and cross disciplinary system engineering. With the continuous emergence of new materials, algorithms, and sensors, drone drive and control technology will inevitably usher in new breakthroughs, providing stronger technical support for the development of low altitude economy and the application of intelligent unmanned systems.
