For autonomous racing, visual servo control represents the optimal solution. However, due to the diversity of external disturbances, the complexity of the competition environment and the risk of flight debugging, visual servo control of racing drones is extremely difficult. To address these issues, this paper aims to propose a novel homogeneous feature extraction-based visual servo control method for accurate racing drones.
A homogeneous feature extraction neural network (HFENet) is proposed to map the different input modalities into an embedding space spanned by the learned homogeneous features representing the inherent structural characteristics. As a result, appropriate control commands are generated through a simple combination of these learned homogeneous features. Additionally, we introduce an optimization method for HFENet that uses imitation learning in conjunction with an expert strategy, and enhance the transferability of the training results from the simulation environment by using domain randomization methods.
The effectiveness of the proposed method is validated through track gate targeting experiments in both simulated and real-world settings. The simulation results demonstrate that, compared to existing methods, the proposed method exhibits superior target alignment performance in complex environments and greater robustness to strong disturbances. Additionally, an indoor flight platform is developed to verify the applicability of the strategies trained in the simulation environment to real-world conditions.
This paper proposes a novel homogeneous feature extraction-based visual servo control method. The proposed method demonstrates superior target alignment performance, enhanced resistance to interference and robust sim-to-real transfer capabilities.
Homogeneous feature extraction-based visual servo control for racing drones
Aircraft Engineering and Aerospace Technology ; 97 , 6 ; 659-667
27.06.2025
9 pages
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Aggressive Trajectory Generation for A Swarm of Autonomous Racing Drones
ArXiv | 2023
|