Climate neutrality objectives require the rapid introduction of novel technologies in the aviation sector, a fact which is set to challenge current aircraft test and certification methods and procedures. In this perspective, this article presents a new methodology for reinforcement-learning-based testing of aircraft handling qualities (HQ), aimed at complementing and enhancing established existent techniques. The proposed method relies on a backpropagation-through-time scheme to calculate optimal control sequences corresponding to state trajectories that, in accordance with test objectives, satisfy or purposely violate output constraints to evaluate selected system limits. Contrary to similar techniques, our methodology includes an explicit treatment of uncertainty, which helps mitigate the effects of the simulation models’ ‘reality gap’: state and output uncertainties are equally propagated in time and used for a probabilistic evaluation of constraint satisfaction, based on the distance from the generated covariance ellipsoids. This approach is shown to produce solutions that are robust to the approximation errors of the models used for training. Finally, the temporal backpropagation method outputs analytic gradients that allow for optimizing control sequences with very high dimensionality at low computational cost. A case study on aircraft HQ testing is presented, focusing on aircraft longitudinal dynamics. In a number of examples and without access to pre-coded information, the algorithm is shown to reproduce well-known test sequences, highlighting its ability to explore high-dimensional input spaces. The latter represents an important asset for improving the future test and validation procedures of novel, unexplored, aircraft configurations.
Reinforcement-learning-based aircraft handling qualities testing with uncertainty modeling
Reinforcement-learning-based aircraft handling qualities testing with uncertainty modeling A. Antonakis
CEAS Aeronaut J
CEAS Aeronautical Journal ; 16 , 2 ; 467-483
2025-04-01
17 pages
Article (Journal)
Electronic Resource
English
Reinforcement-learning-based aircraft handling qualities testing with uncertainty modeling
Springer Verlag | 2025
|TIBKAT | 1999
|Large aircraft handling qualities
NTRS | 1984
|Aircraft handling qualities data
TIBKAT | 1972
|Aircraft Handling Qualities Data
NTIS | 1972
|