Lin-Jie Huang,∗ Zi-Huan Cheng,∗ Hai-Long Pei,∗ Ding Xu,∗ and Zhi-Hao Xu∗∗
System identification, Q-learning, orthonormal basis functions, electric direct-drive coaxial unmanned helicopters (ED-DCUHs)
System modelling is crucial for improving the flight quality of unmanned helicopters. However, the multi-channel coupling and complex aerodynamic characteristics, particularly the aerodynamic interference between rotors in electric direct-drive coaxial unmanned helicopters (ED-DCUHs), pose significant challenges for precise modelling. To address the yaw channel modelling problem of ED-DCUHs, we propose an orthogonal basis pole evaluation system identification algorithm based on improved Q-learning. By introducing orthogonal basis functions, this method resolves the over- parameterisation issue of traditional frequency domain identification methods and utilises prior knowledge to optimise the pole evaluation process. Additionally, by reducing the dimensionality of the Q- value function, the method alleviates the “curse of dimensionality” inherent in Q-learning to some extent. Simulations and flight tests demonstrate the effectiveness and superiority of this approach.
Important Links:
Go Back
