TY - GEN
T1 - A hybrid learning approach for the simulation of unmanned aircraft vehicle dynamics
AU - Yu, Yang
AU - Liu, Yongming
N1 - Funding Information:
The research reported in this paper was supported by funds from NASA University Leadership Initiative program (Contract No. NNX17AJ86A, PI: Yongming Liu, Project Officer: Kai Goebel). The support is gratefully acknowledged.
Publisher Copyright:
© 2019, American Institute of Aeronautics and Astronautics Inc, AIAA. All rights reserved.
PY - 2019
Y1 - 2019
N2 - The computation costs of predicting flight trajectories of unmanned aircraft vehicle (UAV) can be expensive or even prohibitive especially for the large number of UAVs in UAV traffic management (UTM). This study proposes the concept of hybrid learning, a novel approach based on data-driven learning and underlying physics, as a computationally efficient method for the simulation of UAV dynamics. The hybrid learning integrates the underlying physics of UAV systems into learning models such as neural networks to reduce the training and simulation costs. The application of hybrid learning for simulating UAV dynamics is demonstrated using a recently introduced physics-aware network known as the deep residual recurrent neural network (DR-RNN) on a quadcopter model. The UAV dynamics are described using a six degrees-of-freedom model. The DR-RNN is used to predict the response of UAV under arbitrary control inputs. The results show that the DR-RNN can accurately predict UAV responses and has excellent extrapolation capabilities. Moreover, the DR-RNN exhibits superior computation efficiency compared with a classical numerical method, the first-order Runge-Kutta method, highlighting its suitability in serving as surrogate models for UAV systems.
AB - The computation costs of predicting flight trajectories of unmanned aircraft vehicle (UAV) can be expensive or even prohibitive especially for the large number of UAVs in UAV traffic management (UTM). This study proposes the concept of hybrid learning, a novel approach based on data-driven learning and underlying physics, as a computationally efficient method for the simulation of UAV dynamics. The hybrid learning integrates the underlying physics of UAV systems into learning models such as neural networks to reduce the training and simulation costs. The application of hybrid learning for simulating UAV dynamics is demonstrated using a recently introduced physics-aware network known as the deep residual recurrent neural network (DR-RNN) on a quadcopter model. The UAV dynamics are described using a six degrees-of-freedom model. The DR-RNN is used to predict the response of UAV under arbitrary control inputs. The results show that the DR-RNN can accurately predict UAV responses and has excellent extrapolation capabilities. Moreover, the DR-RNN exhibits superior computation efficiency compared with a classical numerical method, the first-order Runge-Kutta method, highlighting its suitability in serving as surrogate models for UAV systems.
UR - http://www.scopus.com/inward/record.url?scp=85098473489&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85098473489&partnerID=8YFLogxK
U2 - 10.2514/6.2019-2940
DO - 10.2514/6.2019-2940
M3 - Conference contribution
AN - SCOPUS:85098473489
SN - 9781624105890
T3 - AIAA Aviation 2019 Forum
BT - AIAA Aviation 2019 Forum
PB - American Institute of Aeronautics and Astronautics Inc, AIAA
T2 - AIAA Aviation 2019 Forum
Y2 - 17 June 2019 through 21 June 2019
ER -