Deep-Reinforcement-Learning-based Nonlinear Adaptive Flight Control
On the gap between simulation and reality
A. Beňo (TU Delft - Aerospace Engineering)
E. van Kampen – Mentor (TU Delft - Aerospace Engineering)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
This paper presents a novel corrective algorithm bridging the gap between simulation and reality by online fine-tuning an offline pre-trained deep reinforcement learning agent. The novel control architecture is inspired by the incremental model-based heuristic dynamic programming, which is described together with the basics of reinforcement learning first. This novel control architecture is applied in an illustrative control environment. It was found that the corrective algorithm can help reach the desired reference state in an environment governed by moderately different dynamics from those used during pre-training of the reinforcement learning agent.
help_outline