Results for “Adaptive learning” — Page 3

NASA NTRS2018

Practical Aspects of Real-Time Modeling for the Learn-To-Fly Concept

Practical aspects of autonomous onboard real-time modeling for the NASA Learn-to-Fly concept are examined using flight data from two subscale test aircraft. A practical real-time global nonlinear aerodynamic modeling method is developed and explained, along with the multiple-input excitation needed for effective real-time modeling. Critical issues for integrating real-time global nonlinear aerodynamic modeling and local linear modeling with the real-time learning adaptive control and guidance elements of the Learn-to-Fly algorithm are identified and discussed. Real-time modeling results from NASA Learn-to-Fly flight demonstrations are presented and evaluated using model fit metrics and prediction tests.

Morelli, Eugene A.↗

NASA NTRS1975

Adaptive Bayes classifiers for remotely sensed data

An algorithm is developed for a learning, adaptive, statistical pattern classifier for remotely sensed data. The estimation procedure consists of two steps: (1) an optimal stochastic approximation of the parameters of interest, and (2) a projection of the parameters in time and space. The results reported are for Gaussian data in which the mean vector of each class may vary with time or position after the classifier is trained.

Raulston, H. S.↗

NASA NTRS1990

Neural controller for adaptive movements with unforeseen payloads

A theory and computer simulation of a neural controller that learns to move and position a link carrying an unforeseen payload accurately are presented. The neural controller learns adaptive dynamic control from its own experience. It does not use information about link mass, link length, or direction of gravity, and it uses only indirect uncalibrated information about payload and actuator limits. Its average positioning accuracy across a large range of payloads after learning is 3 percent of the positioning range. This neural controller can be used as a basis for coordinating any number of sensory inputs with limbs of any number of joints. The feedforward nature of control allows parallel implementation in real time across multiple joints.

Kuperstein, Michael↗

NASA NTRS1967

Learning models for adaptive system identification, part A Final report

Learning models for identification, adaptation, and sensitivity of optimal control systems

Sherman, J. W.↗

DOE OSTI2025

Deep-learning-based domain adaptation for cavity fault prediction at Jefferson Laboratory

Superconducting radio-frequency (SRF) cavities are the core components of the Continuous Electron Beam Accelerator Facility (CEBAF) at Jefferson Lab, providing high-power electron beams for nuclear physics experiments. The facility comprises 418 SRF cavities, and any fault in these cavities can lead to interruptions in the electron beam supply. Cavity faults are the leading cause of beam trips in CEBAF. Predicting and mitigating those faults before onset can help maintain normal operation. Existing models face challenges in distinguishing between normal and fault signals when changes occur in the underlying time-series data, from changes in control software, operational parameters, or the environment. This work proposes a deep learning domain adaptation model that leverages transfer learning to address fault prediction challenges by improving accuracy. The model is trained and fine-tuned using a dataset collected for faulty and normal operation using a data acquisition system in CEBAF. Our deep learning-based domain adaptation model achieves a prediction accuracy of 89.61% of the fault and normal signals. The developed model effectively predicts normal running signals compared to the baseline approach without domain adaptation. This capacity is essential for the fault prediction task in the CEBAF because of heavily imbalanced data containing vast amounts of normal signals. The model performs well for predicting faults several hundred milliseconds before the fault onset compared to other models where no adaptation is applied. Incorporating deep learning-based domain adaptation techniques will significantly improve the fault prediction performance.

Rahman, Md Monibor [Old Dominion Univ., Norfolk, V↗

NASA NTRS2018

NASA's Learn-to-Fly Project Overview

Learn-to-Fly (L2F) is an advanced technology development effort aimed at assessing the feasibility of real-time, self-learning flight vehicles. Specifically, research has been conducted on merging real-time aerodynamic modeling, learning adaptive control, and other disciplines with the goal of using this “learn to fly” methodology to replace the current iterative vehicle development paradigm, substantially reducing the typical ground and flight testing requirements for air vehicle design. Recent activities included an aggressive flight test program with unique fully autonomous fight test vehicles to rapidly advance L2F technology. This paper presents an overview of the project and key components.

Heim, Eugene H. D.↗

NASA NTRS2021

Safe Tracking Control of an Uncertain Euler-Lagrange System with Full-State Constraints using Barrier Functions

This paper presents a novel, safe tracking control design method that learns the parameters of an uncertain Euler-Lagrange (EL) system online using adaptive learning laws. A barrier function (BF) is first used to transform the full-state constrained EL-dynamics into an equivalent unconstrained dynamics. An adaptive tracking controller is then developed along with the parameter update law in the transformed state space such that the states remain bounded for all time within a prescribed bound. A stability analysis is developed that considers the EL-dynamics’ uncertainty, yielding a semi-globally uniformly ultimately bounded (SGUUB) tracking error and the parameter estimation error. The controller design is validated in simulations using a two-link planar manipulator. The results show the proposed method’s ability to track the reference trajectory while remaining inside each of the predefined state bounds.

Robots↗

NASA NTRS2017

TPSAS-NF1676L-28077-DND

Learn-To-Fly merges real-time modeling, learning adaptive control, and real-time guidance to achieve: (1) autonomous self-learning aircraft to disrupt our current vehicle development paradigm, (2) novel configuration development with no ground testing, and (3) robust autonomous flight of novel aircraft.

Mark A Croom↗

NASA NTRSDate not supplied

Discovery and Analysis of Rare High-Impact Failure Modes using Adversarial RL-Informed Sampling

Adaptive learning agents have tremendous potential to handle critical tasks currently performed by humans. Unfortunately, due to their complexity, it can be difficult to verify that these learning agents do not have critical failure modes. Standard verification and validation methods often do not apply directly to learning agents and Monte Carlo methods have difficulty covering even a small fraction of the state space, especially in multiagent systems or over long time horizons. To overcome this difficulty, we demonstrate an adaptive stress-testing method based on reinforcement learning of correlations that raise the probability of failure. This approach has three key properties: (1) it is able to find rare failure modes with far greater sample efficiency than Monte Carlo methods, (2) it can estimate the true probability of a failure mode despite the inherent bias in the learning method, and (3) it is capable of learning and resampling compact representations of multimodal failure spaces. These properties are important in practice as we need to find disparate failure modes while accounting for their actual relevance. This is a significant advantage over traditional adaptive stress testing methods that give abstract likelihoods of particular failure instances, but cannot estimate the probability of a broader failure mode. We test our algorithm on a simple problem from the aviation domain where an autonomous aircraft lands in gusty wind conditions. The results suggest that we can find failure modes with far fewer samples than the Monte Carlo approach and simultaneously estimate the probability of failure.

reinforcement learning↗

NASA NTRSDate not supplied

Discovery and Analysis of Rare High-Impact Failure Modes using Adversarial RL-Informed Sampling

Adaptive learning agents have tremendous potential to handle critical tasks currently performed by humans. Unfortunately, due to their complexity, it can be difficult to verify that these learning agents do not have critical failure modes. Standard verification and validation methods often do not apply directly to learning agents and Monte Carlo methods have difficulty covering even a small fraction of the state space, especially in multiagent systems or over long time horizons. To overcome this difficulty, we demonstrate an adaptive stress-testing method based on reinforcement learning of correlations that raise the probability of failure. This approach has three key properties: (1) it is able to find rare failure modes with far greater sample efficiency than Monte Carlo methods, (2) it can estimate the true probability of a failure mode despite the inherent bias in the learning method, and (3) it is capable of learning and resampling compact representations of multimodal failure spaces. These properties are important in practice as we need to find disparate failure modes while accounting for their actual relevance. This is a significant advantage over traditional adaptive stress testing methods that give abstract likelihoods of particular failure instances, but cannot estimate the probability of a broader failure mode. We test our algorithm on a simple problem from the aviation domain where an autonomous aircraft lands in gusty wind conditions. The results suggest that we can find failure modes with far fewer samples than the Monte Carlo approach and simultaneously estimate the probability of failure.

Validation↗

NASA NTRSDate not supplied

Discovery and Analysis of Rare High-Impact Failure Modes using Adversarial RL-Informed Sampling

Adaptive learning agents have tremendous potential to handle critical tasks currently performed by humans. Unfortunately, due to their complexity, it can be difficult to verify that these learning agents do not have critical failure modes. Standard verification and validation methods often do not apply directly to learning agents and Monte Carlo methods have difficulty covering even a small fraction of the state space, especially in multiagent systems or over long time horizons. To overcome this difficulty, we demonstrate an adaptive stress-testing method based on reinforcement learning of correlations that raise the probability of failure. This approach has three key properties: (1) it is able to find rare failure modes with far greater sample efficiency than Monte Carlo methods, (2) it can estimate the true probability of a failure mode despite the inherent bias in the learning method, and (3) it is capable of learning and resampling compact representations of multimodal failure spaces. These properties are important in practice as we need to find disparate failure modes while accounting for their actual relevance. This is a significant advantage over traditional adaptive stress testing methods that give abstract likelihoods of particular failure instances, but cannot estimate the probability of a broader failure mode. We test our algorithm on a simple problem from the aviation domain where an autonomous aircraft lands in gusty wind conditions. The results suggest that we can find failure modes with far fewer samples than the Monte Carlo approach and simultaneously estimate the probability of failure.

Validation↗

NASA NTRS2004

Monitoring the Performance of a Neuro-Adaptive Controller

Traditional control has proven to be ineffective to deal with catastrophic changes or slow degradation of complex, highly nonlinear systems like aircraft or spacecraft, robotics, or flexible manufacturing systems. Control systems which can adapt toward changes in the plant have been proposed as they offer many advantages (e.g., better performance, controllability of aircraft despite of a damaged wing). In the last few years, use of neural networks in adaptive controllers (neuro-adaptive control) has been studied actively. Neural networks of various architectures have been used successfully for online learning adaptive controllers. In such a typical control architecture, the neural network receives as an input the current deviation between desired and actual plant behavior and, by on-line training, tries to minimize this discrepancy (e.g.; by producing a control augmentation signal). Even though neuro-adaptive controllers offer many advantages, they have not been used in mission- or safety-critical applications, because performance and safety guarantees cannot b e provided at development time-a major prerequisite for safety certification (e.g., by the FAA or NASA). Verification and Validation (V&V) of an adaptive controller requires the development of new analysis techniques which can demonstrate that the control system behaves safely under all operating conditions. Because of the requirement to adapt toward unforeseen changes during operation, i.e., in real time, design-time V&V is not sufficient.

Schumann, Johann↗

NASA NTRS2002

Method and system for training dynamic nonlinear adaptive filters which have embedded memory

Described herein is a method and system for training nonlinear adaptive filters (or neural networks) which have embedded memory. Such memory can arise in a multi-layer finite impulse response (FIR) architecture, or an infinite impulse response (IIR) architecture. We focus on filter architectures with separate linear dynamic components and static nonlinear components. Such filters can be structured so as to restrict their degrees of computational freedom based on a priori knowledge about the dynamic operation to be emulated. The method is detailed for an FIR architecture which consists of linear FIR filters together with nonlinear generalized single layer subnets. For the IIR case, we extend the methodology to a general nonlinear architecture which uses feedback. For these dynamic architectures, we describe how one can apply optimization techniques which make updates closer to the Newton direction than those of a steepest descent method, such as backpropagation. We detail a novel adaptive modified Gauss-Newton optimization technique, which uses an adaptive learning rate to determine both the magnitude and direction of update steps. For a wide range of adaptive filtering applications, the new training algorithm converges faster and to a smaller value of cost than both steepest-descent methods such as backpropagation-through-time, and standard quasi-Newton methods. We apply the algorithm to modeling the inverse of a nonlinear dynamic tracking system 5, as well as a nonlinear amplifier 6.

Rabinowitz, Matthew↗

DOE OSTI2025

Criticality analysis of nuclear binding energy neural networks

Machine learning methods, in particular deep learning methods such as artificial neural networks (ANNs) with many layers, have become widespread and useful tools in nuclear physics. However, these ANNs are typically treated as ‘black boxes’, with their architecture (width, depth, and weight/bias initialization) and the training algorithm and parameters chosen empirically by optimizing learning based on limited exploration. We test a non-empirical approach to understanding and optimizing nuclear physics ANNs by adapting a criticality analysis based on renormalization group flows in terms of the hyperparameters for weight/bias initialization, training rates, and the ratio of depth to width. This treatment utilizes the statistical properties of neural network initialization to find a generating functional for network outputs at any layer, allowing for a path integral formulation of the ANN outputs as a Euclidean statistical field theory. We use a prototypical example to test the applicability of this approach: a simple ANN for nuclear binding energies. We find that with training using a stochastic gradient descent optimizer, the predicted criticality behavior is realized, and optimal performance is found with critical tuning. However, the use of an adaptive learning algorithm leads to somewhat superior results without concern for tuning and thus obscures the analysis. Nevertheless, the criticality analysis offers a way to look within the black box of ANNs, which is a first step towards potential improvements in network performance beyond using adaptive optimizers.

artificial neural network↗

NASA NTRS2012

Locomotor Dysfunction after Long-Duration Space Flight and Development of Countermeasures to Facilitate Faster Recovery

Exposure to the microgravity conditions of space flight induces adaptive modification in sensorimotor function allowing astronauts to operate in this unique environment. This adaptive state, however, is inappropriate for a 1-g environment. Consequently astronauts must spend time readapting to Earth s gravity following their return to Earth. During this readaptation period, alterations in sensorimotor function cause various disturbances in astronaut gait during postflight walking. They often rely more on vision for postural and gait stability and many report the need for greater cognitive supervision of motor actions that previous to space flight were fully automated. Over the last several years our laboratory has investigated postflight astronaut locomotion with the aim of better understanding how adaptive changes in underlying sensorimotor mechanisms contribute to postflight gait dysfunction. Exposure to the microgravity conditions of space flight induces adaptive modification in the control of vestibularly-mediated reflexive head movement during locomotion after space flight. Furthermore, during motor learning, adaptive transitions are composed of two main mechanisms: strategic and plastic. Strategic mechanisms represent immediate and transitory modifications in control to deal with changes in the prevailing environment that, if prolonged, induce plastic mechanisms designed to automate new behavioral responses. The goal of the present study was to examine the contributions of sensorimotor subsystems such as the vestibular and body load sensing (BLS) somatosensory influences on head movement control during locomotion after long-duration space flight. Further we present data on the two motor learning processes during readaptation of locomotor function after long-duration space flight.

Mulavara, A. P.↗

NASA NTRS1999

Context-Specific Adaptation of Gravity-Dependent Vestibular Reflex Responses

Stabilization of the eyes and head during body movements is important for maintaining balance and keeping the images of objects stationary on our retinas. Impairment of this ability can lead to disorientation and reduced performance in sensorimotor tasks such as piloting of spacecraft. In the absence of a normal earth gravity field, the dynamics of head stabilization, and the interpretation of vestibular signals that sense gravity and linear acceleration, are subject to change. Transitions between different gravitoinertial force environments - as during different phases of space flight - provide an extreme test of the adaptive mechanisms that maintain these reflexive abilities. It is vitally important to determine human adaptive capabilities in such a circumstance, so that we can know to what extent the sensorimotor skills acquired in one gravity environment will transfer to others. Our work lays the foundation for understanding these capabilities, and for determining how we can aid the processes of adaptation and readaptation. An integrated set of experiments addresses this issue. We use the general approach of adapting some type of reflexive eye movement (saccades, the angular vestibulo-ocular reflex (AVOR), the linear vestibulo-ocular reflex (LVOR)), or the vestibulo-collic reflex (VCR), to a particular change in gain or phase in one condition of gravitoiner-tial force, and adapting to a different gain or phase (or asking for no change) in a second gravitoinertial force condition, and then seeing if the gravitoinertial force itself - the context cue - can recall the previously learned adapted responses. The majority of the experiments in the laboratory use the direction of vertical gaze or the direction of gravity (head tilt) as the context cue. This allows us to study context-specificity in a ground-based setting. One set of experiments, to be performed in parabolic flight, specifically uses the magnitude of gravitoinertial force as a context cue. This is a much better analog of the situation encountered in space flight. Various experiments investigate the behavioral properties, neurophysiological basis, and anatomical substrate of context-specific learning mechanisms. We use otolith (gravity) signals as the contextual cue for switching between adapted states of the saccadic system, the angular and linear vestibulo-ocular reflexes, and the VCR. (By LVOR we mean the oculomotor response - horizontal, vertical, and torsional - to linear translation of the head and body.) We are studying the effect of context on adaptation of saccade gain, phase and gain of the AVOR and LVOR, on ocular counterrolling (OCR) in response to static head tilt, and on head/neck reflexes (VCR) in response to rotation in different orientations. Such research is particularly germane to potential problems of postural and oculomotor control upon exposure to different gravitational environments.

Shelhamer, Mark J.↗

NASA NTRS1999

Context-Specific Adaptation of Gravity-Dependent Vestibular Reflex Responses (NSBRI Neurovestibular Project 1)

Impairment of gaze and head stabilization reflexes can lead to disorientation and reduced performance in sensorimotor tasks such as piloting of spacecraft. Transitions between different gravitoinertial force (gif) environments - as during different phases of space flight - provide an extreme test of the adaptive capabilities of these mechanisms. We wish to determine to what extent the sensorimotor skills acquired in one gravity environment will transfer to others, and to what extent gravity serves as a context cue for inhibiting such transfer. We use the general approach of adapting a response (saccades, vestibuloocular reflex: VOR, or vestibulocollic reflex: VCR) to a particular change in gain or phase in one gif condition, adapting to a different gain or phase in a second gif condition, and then seeing if gif itself - the context cue - can recall the previously-learned adapted responses. Previous evidence indicates that unless there is specific training to induce context-specificity, reflex adaptation is sequential rather than simultaneous. Various experiments in this project investigate the behavioral properties, neurophysiological basis, and anatomical substrate of context-specific learning, using otolith (gravity) signals as a context cue. In the following, we outline the methods for all experiments in this project, and provide details and results on selected experiments.

Shelhamer, Mark↗

NASA NTRS1993

Neural-network-directed alignment of optical systems using the laser-beam spatial filter as an example

This report describes an effort at NASA Lewis Research Center to use artificial neural networks to automate the alignment and control of optical measurement systems. Specifically, it addresses the use of commercially available neural network software and hardware to direct alignments of the common laser-beam-smoothing spatial filter. The report presents a general approach for designing alignment records and combining these into training sets to teach optical alignment functions to neural networks and discusses the use of these training sets to train several types of neural networks. Neural network configurations used include the adaptive resonance network, the back-propagation-trained network, and the counter-propagation network. This work shows that neural networks can be used to produce robust sequencers. These sequencers can learn by example to execute the step-by-step procedures of optical alignment and also can learn adaptively to correct for environmentally induced misalignment. The long-range objective is to use neural networks to automate the alignment and operation of optical measurement systems in remote, harsh, or dangerous aerospace environments. This work also shows that when neural networks are trained by a human operator, training sets should be recorded, training should be executed, and testing should be done in a manner that does not depend on intellectual judgments of the human operator.

Decker, Arthur J.↗