A helicopter is a highly nonlinear system. Its mathematical model is difficult to establish accurately, especially the complicated flapping dynamics. In addition, the forces and moments exerted on the fuselage are very vulnerable to external disturbances like wind gust when flying in the outdoor environment. This paper proposes a composite control scheme which consists of a nonlinear backstepping controller and an extended state observer (ESO) to handle the above problems. The stability of the closed-loop system can be guaranteed based on Lyapunov theory. The external disturbances and model nonlinearities are treated as a lumped disturbance. Meanwhile, the ESO is employed to compensate the influence by estimating the lumped disturbance in real-time. Numerical simulation results are presented to demonstrate that the algorithm can achieve accurate and agile attitude tracking under the external wind gust disturbances even with model uncertainties. When coming to the flight test, a block dropping device was designed to generate a quantifiable and replicable disturbance, and the experimental results indicate that the algorithm introduced above can reject the external disturbance rapidly and track the given attitude command precisely.

Active tracking control is essential for UAVs to perform autonomous operations in GPS-denied environments. In the active tracking task, UAVs take high-dimensional raw images as input and execute motor actions to actively follow the dynamic target. Most research focuses on three-stage methods, which entail perception first, followed by high-level decision-making based on extracted spatial information of the dynamic target, and then UAV movement control, using a low-level dynamic controller. Perception methods based on deep neural networks are powerful but require considerable effort for manual ground truth labeling. Instead, we unify the perception and decision-making stages using a high-level controller and then leverage deep reinforcement learning to learn the mapping from raw images to the high-level action commands in the V-REP-based environment, where simulation data are infinite and inexpensive. This end-to-end method also has the advantages of a small parameter size and reduced effort requirements for parameter turning in the decision-making stage. The high-level controller, which has a novel architecture, explicitly encodes the spatial and temporal features of the dynamic target. Auxiliary segmentation and motion-in-depth losses are introduced to generate denser training signals for the high-level controller’s fast and stable training. The high-level controller and a conventional low-level PID controller constitute our hierarchical active tracking control framework for the UAVs’ active tracking task. Simulation experiments show that our controller trained with several augmentation techniques sufficiently generalizes dynamic targets with random appearances and velocities, and achieves significantly better performance, compared with three-stage methods.

Target detection and tracking can be widely used in military and civilian scenarios. Unmanned aerial vehicles (UAVs) have high maneuverability and strong concealment, thus they are very suitable for using as a platform for ground target detection and tracking. Most of the existing target detection and tracking algorithms are aimed at conventional targets. Because of the small scale and the incomplete details of the targets in the aerial image, it is difficult to apply the conventional algorithms to aerial photography from UAVs. This paper proposes a ground target image detection and tracking algorithm applied to UAVs using a revised deep learning technology. Aiming at the characteristics of ground targets in aerial images, target detection algorithms and target tracking algorithms are improved. The target detection algorithm is improved to detect small targets on the ground. The target tracking algorithm is designed to recover the target after the target is lost. The target detection and tracking algorithm is verified on the aerial dataset.

In order to fly safely and autonomously in complex environments, UAVs need to be able to plan their trajectories in real-time. This paper proposes an improved B-spline-based trajectory generation method that can generate safe, smooth, and kinodynamically feasible trajectories in real-time. This paper firstly introduces the principle of error upper bound of the B-spline curve and proposes a new trajectory safety assurance method; then, the loss function of trajectory is constructed based on safety, smoothness, and flight time; finally, a parameter adaptive trajectory optimization method is proposed, so that obtain the safe trajectory. Compared with the existing methods, the proposed method has two important improvements: (1) it solves the problem of overly conservative safety distance estimation at control points, improves the trajectory smoothness, and reduces the required flight time; (2) it proposes a trajectory optimization method with adaptive adjustment of safety distance parameters, which improves the quality and success rate of the planned trajectory. We validate our proposed method in simulation and real-world tasks, and the test results show that the method proposed in this paper can significantly improve the quality of the generated trajectory.