Hierarchical Reinforcement Learning and Gain Scheduling-based Control of a Hypersonic Vehicle

Makumi, Wanjiku A.; Greene, Max L.; Bell, Zachary I.; Bialy, Brendan; Kamalapurkar, Rushikesh; Dixon, Warren E.

doi:10.2514/6.2023-2505

Cited by 1 publication

(1 citation statement)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For instance, autonomous underwater vehicles and exoskeleton systems have complex and uncertain dynamics and thus difficult to model accurately [2]. Alternatively, the structure of the system dynamics may be known in some application scenarios, but the model parameters may be unknown or uncertain, as in the case of hypersonic vehicles [16]. When faced with the lack of system information, identifier/observer-based reconstruction methods have been employed to facilitate the ADP framework.…”

Section: Introductionmentioning

confidence: 99%

Event-triggered optimal control of completely unknown nonlinear systems via identifier-critic learning

Peng¹,

Zhang²,

Luo³

et al. 2023

Kybernetika

View full text Add to dashboard Cite

This paper proposes an online identifier-critic learning framework for event-triggered optimal control of completely unknown nonlinear systems. Unlike classical adaptive dynamic programming (ADP) methods with actor-critic neural networks (NNs), a filter-regression-based approach is developed to reconstruct the unknown system dynamics, and thus avoid the dependence on an accurate system model in the control design loop. Meanwhile, NN adaptive laws are designed for the parameter estimation by using only the measured system state and input data, and facilitate the identifier-critic NN design. The convergence of the adaptive laws is analyzed. Furthermore, in order to reduce state sampling frequency, two kinds of aperiodic sampling schemes, namely static and dynamic event triggers, are embedded into the proposed optimal control design. Finally, simulation results are presented to demonstrate the effectiveness of the proposed event-triggered optimal control strategy.

show abstract

Section: Introductionmentioning

confidence: 99%