Improving Long-Term Learning of Model Reference Adaptive Controllers for Flight Applications: A Sparse Neural Network Approach

Nivison, Scott; Khargonekar, Pramod P.

doi:10.2514/6.2017-1249

Cited by 15 publications

(9 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The goal of this research is to further develop a recently conceptualized novel adaptive control architecture called sparse neural network (SNN) adaptive control (see Ref. [6]) and demonstrate the capabilities of the SNN by controlling a HSV with flexible body effects. The use of an adaptive controller is necessary because of the highly nonlinear and time-varying nature of the flight vehicle dynamics as well as the existence of significant unmodeled dynamics.…”

Section: Introductionmentioning

confidence: 99%

“…Inspired by the success of distributed sparse neural networks in the machine learning community, the SNN adaptive control architecture was originally conceived to improve controller memory and tracking performance of flight vehicles with practical processor limitations that encounter persistent significant uncertainties [6]. The SNN succeeds by segmenting the flight envelope, increasing the overall number of neurons available to the adaptive system, and only selecting a small percentage of those neurons to be active in the adaptive controller based on location within the flight envelope.…”

Section: Introductionmentioning

confidence: 99%

“…The additional neural network memory allows the SNN controller to store and recall neural network weights and uncertainty estimates from each segmented region in the flight envelope. Hence, the beneficial structure allows additional performance benefits over a traditional single hidden layer (SHL) adaptive controller [6].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A Sparse Neural Network Approach to Model Reference Adaptive Control with Hypersonic Flight Applications

Nivison

Khargonekar

2018

2018 AIAA Guidance, Navigation, and Control Conference

Self Cite

View full text Add to dashboard Cite

Neural network-based model reference adaptive control (MRAC) is an effective architecture used in the flight control community to combat significant uncertainties where the structure of the uncertainty is unknown. In our previous work, a novel adaptive control architecture called sparse neural network (SNN) was developed in order to improve long-term learning and transient performance of flight vehicles with persistent uncertainties which lie in various regions throughout the operating regime. The SNN is designed to operate with small learning rates in order to avoid high-frequency oscillations and utilizes only a small number of active neurons in the adaptive controller in order to reduce the computational burden on the processor. In this paper, we enhance the SNN architecture by developing an innovative adaptive control term which is used to mitigate a control effectiveness matrix. Furthermore, we design a robust control term and a strict dwell time condition in order to ensure stability while switching between segments. We demonstrate the effectiveness of the SNN approach by controlling a sophisticated hypersonic vehicle with flexible body effects.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Sparse Neural Network Approach to Model Reference Adaptive Control with Hypersonic Flight Applications

Nivison

Khargonekar

2018

2018 AIAA Guidance, Navigation, and Control Conference

Self Cite

View full text Add to dashboard Cite

show abstract

“…The developed generalization is motivated by adaptive control of switched systems where the derivative of the candidate Lyapunov function is typically negative semidefinite.Index Terms switched systems, differential inclusions, adaptive systems, nonlinear systems Rushikesh Kamalapurkar is with the School 2 benefit from adaptive methods where the controller adapts to the uncertain dynamics without strictly relying on high gain or high frequency feedback often associated with robust control methods that can lead to overstimulation.Switched dynamics are inherent in a variety of modern adaptation strategies. For example, in sparse neural networks [5], the use of different approximation architectures for different regions of the state-space introduce switching via the feedforward part of the controller. In adaptive gain scheduling methods [6], switching is introduced due to changing feedback gains.…”

mentioning

confidence: 99%

“…Switched dynamics are inherent in a variety of modern adaptation strategies. For example, in sparse neural networks [5], the use of different approximation architectures for different regions of the state-space introduce switching via the feedforward part of the controller. In adaptive gain scheduling methods [6], switching is introduced due to changing feedback gains.…”

mentioning

confidence: 99%

Invariance-Like Results for Nonautonomous Switched Systems

Kamalapurkar

Rosenfeld

Parikh

et al. 2019

IEEE Trans. Automat. Contr.

View full text Add to dashboard Cite

This paper generalizes the LaSalle-Yoshizawa Theorem to switched nonsmooth systems. Filippov and Krasovskii regularizations of a switched system are shown to be contained within the convex hull of the Filippov and Krasovskii regularizations of the subsystems, respectively. A candidate common Lyapunov function that has a negative semidefinite derivative along the trajectories of the subsystems is shown to be sufficient to establish LaSalle-Yoshizawa results for the switched system. Results for regular and non-regular candidate Lyapunov functions are developed via appropriate generalization of the notion of a time derivative. The developed generalization is motivated by adaptive control of switched systems where the derivative of the candidate Lyapunov function is typically negative semidefinite.Index Terms switched systems, differential inclusions, adaptive systems, nonlinear systems Rushikesh Kamalapurkar is with the School 2 benefit from adaptive methods where the controller adapts to the uncertain dynamics without strictly relying on high gain or high frequency feedback often associated with robust control methods that can lead to overstimulation.Switched dynamics are inherent in a variety of modern adaptation strategies. For example, in sparse neural networks [5], the use of different approximation architectures for different regions of the state-space introduce switching via the feedforward part of the controller. In adaptive gain scheduling methods [6], switching is introduced due to changing feedback gains. Switching is also utilized as a tool to improve transient response of adaptive controllers by selecting between multiple estimated models of stable plants (see, e.g., [7]-[16]). In addition, switched systems theory can be utilized to extend the scope of existing adaptive solutions to more complex circumstances that involve switched dynamics.Lyapunov-based stability analysis of switched nonautonomous adaptive systems is challenging because adaptive update laws typically result in non-strict Lyapunov functions for the individual subsystems. For each subsystem, convergence of the error signal to the origin is typically established using Barbȃlat's lemma (e.g., [17, Lemma 8.2]).In traditional methods that utilize multiple Lyapunov functions (e.g., [18, Theorem 3.2]) the class of admissible switching signals is restricted based on the rate of decay of the cLf (cf. [18, Eq. 3.10]). Since Barbȃlat's lemma provides no information about the rate of decay of the cLf, it alone is insufficient to establish stability of a switched system using multiple Lyapunov functions. Approaches based on common cLfs have been developed for systems with negative definite Lyapunov derivatives; however, common cLf-based approaches do not trivially extend to systems with non-strict Lyapunov functions (cf. [19]-[21] and [18, Example 2.1]). Because of complications resulting from a negative semidefinite Lyapunov derivative, few results are available in literature that examine adaptive control of uncertain nonlinear switched systems. An...

show abstract