Results for “practical reinforcement learning”

NASA NTRS1999

Reinforcement Learning in a Nonstationary Environment: The El Farol Problem

This paper examines the performance of simple learning rules in a complex adaptive system based on a coordination problem modeled on the El Farol problem. The key features of the El Farol problem are that it typically involves a medium number of agents and that agents' pay-off functions have a discontinuous response to increased congestion. First we consider a single adaptive agent facing a stationary environment. We demonstrate that the simple learning rules proposed by Roth and Er'ev can be extremely sensitive to small changes in the initial conditions and that events early in a simulation can affect the performance of the rule over a relatively long time horizon. In contrast, a reinforcement learning rule based on standard practice in the computer science literature converges rapidly and robustly. The situation is reversed when multiple adaptive agents interact: the RE algorithms often converge rapidly to a stable average aggregate attendance despite the slow and erratic behavior of individual learners, while the CS based learners frequently over-attend in the early and intermediate terms. The symmetric mixed strategy equilibria is unstable: all three learning rules ultimately tend towards pure strategies or stabilize in the medium term at non-equilibrium probabilities of attendance. The brittleness of the algorithms in different contexts emphasize the importance of thorough and thoughtful examination of simulation-based results.

Bell, Ann Maria↗

NASA NTRS1992

Fuzzy logic and neural network technologies

Applications of fuzzy logic technologies in NASA projects are reviewed to examine their advantages in the development of neural networks for aerospace and commercial expert systems and control. Examples of fuzzy-logic applications include a 6-DOF spacecraft controller, collision-avoidance systems, and reinforcement-learning techniques. The commercial applications examined include a fuzzy autofocusing system, an air conditioning system, and an automobile transmission application. The practical use of fuzzy logic is set in the theoretical context of artificial neural systems (ANSs) to give the background for an overview of ANS research programs at NASA. The research and application programs include the Network Execution and Training Simulator and faster training algorithms such as the Difference Optimized Training Scheme. The networks are well suited for pattern-recognition applications such as predicting sunspots, controlling posture maintenance, and conducting adaptive diagnoses.

Villarreal, James A.↗

NASA NTRS1998

Preliminary Work for Examining the Scalability of Reinforcement Learning

Researchers began studying automated agents that learn to perform multiple-step tasks early in the history of artificial intelligence (Samuel, 1963; Samuel, 1967; Waterman, 1970; Fikes, Hart & Nilsonn, 1972). Multiple-step tasks are tasks that can only be solved via a sequence of decisions, such as control problems, robotics problems, classic problem-solving, and game-playing. The objective of agents attempting to learn such tasks is to use the resources they have available in order to become more proficient at the tasks. In particular, each agent attempts to develop a good policy, a mapping from states to actions, that allows it to select actions that optimize a measure of its performance on the task; for example, reducing the number of steps necessary to complete the task successfully. Our study focuses on reinforcement learning, a set of learning techniques where the learner performs trial-and-error experiments in the task and adapts its policy based on the outcome of those experiments. Much of the work in reinforcement learning has focused on a particular, simple representation, where every problem state is represented explicitly in a table, and associated with each state are the actions that can be chosen in that state. A major advantage of this table lookup representation is that one can prove that certain reinforcement learning techniques will develop an optimal policy for the current task. The drawback is that the representation limits the application of reinforcement learning to multiple-step tasks with relatively small state-spaces. There has been a little theoretical work that proves that convergence to optimal solutions can be obtained when using generalization structures, but the structures are quite simple. The theory says little about complex structures, such as multi-layer, feedforward artificial neural networks (Rumelhart & McClelland, 1986), but empirical results indicate that the use of reinforcement learning with such structures is promising. These empirical results make no theoretical claims, nor compare the policies produced to optimal policies. A goal of our work is to be able to make the comparison between an optimal policy and one stored in an artificial neural network. A difficulty of performing such a study is finding a multiple-step task that is small enough that one can find an optimal policy using table lookup, yet large enough that, for practical purposes, an artificial neural network is really required. We have identified a limited form of the game OTHELLO as satisfying these requirements. The work we report here is in the very preliminary stages of research, but this paper provides background for the problem being studied and a description of our initial approach to examining the problem. In the remainder of this paper, we first describe reinforcement learning in more detail. Next, we present the game OTHELLO. Finally we argue that a restricted form of the game meets the requirements of our study, and describe our preliminary approach to finding an optimal solution to the problem.

Clouse, Jeff↗

NASA NTRSDate not supplied

Structural Carbon Nanotube Development

Mechanical properties of carbon nanotubes (CNTs) at the nanoscale suggest their potential to enable significantly lighter structures of interest in aerospace applications. As different formats of the material became available over the last 20 years, understanding of how CNTs contribute to improvements in mechanical properties of structural composites has evolved.Although the material is now available in large quantities that permit the development of composite fabrication processes, the retention of properties that make these materials attractive on the nano scale remains a challenge in large assemblies of CNTs. The work to be presented will summarize lessons learned in our efforts to produce carbon nanotube composites with tensile properties that begin to rival those of carbon fiber reinforced polymer composites. Challenges that need to be overcome for practical adoption of this emerging material system will be considered.

Carbon nanotube↗

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 NTRS1995

Indicator Systems and Evaluation

Participants in the workshop session were actively engaged in a hands-on, minds-on approach to learning about indicators and evaluation processes. The six hour session was broken down into three two hour sessions. Each session was built upon an instructional model which moved from general understanding to specific IITA application. Examples and practice exercises served to demonstrate tand reinforce the workshop concepts. Each successive session built upon the previous session and addressed the major steps in the evaluation process. The major steps covered in the workshop included: project descriptions, writing goals and objectives for categories, determining indicators and indicator systems for specific projects, and methods and issues of data collection. The workshop served as a baseline upon which the field centers will build during the summer in undertaking a comprehensive examination and evaluation of their existing K-12 education projects.

Canright, Shelley↗

NASA NTRS2022

SatNet: A Benchmark for Satellite Scheduling Optimization

Satellites provide essential services such as networking and weather tracking, and the number of near-earth and deep space satellites are expected to grow rapidly in the coming years. Communications with terrestrial ground stations is one of the critical functionalities of any space mission. Satellite scheduling is a problem that has been scientifically investigated since the 1970s. A central aspect of this problem is the need to consider resource contention and satellite visibility constraints as they require line of sight. Due to the combinatorial nature of the problem, prior solutions such as linear programs and evolutionary algorithms require extensive compute capabilities to output a feasible schedule for each scenario. Machine learning based scheduling can provide an alternative solution by training a model with historical data and generating a schedule quickly with model inference. We present SatNet, a benchmark for satellite scheduling optimization based on historical data from the NASA Deep Space Network. We propose formulation of the satellite scheduling problem as a Markov Decision Process and use reinforcement learning (RL) policies to generate schedules. The nature of constraints imposed by SatNet differ from other combinatorial optimization problems such as vehicle routing studied in prior literature. Our initial results indicate that RL is an alternative optimization approach that can generate candidate solutions of comparable quality to existing state-of-the-practice results. However, we also find that RL policies overfit to the training dataset and do not generalize well to new data, thereby necessitating continued research on reusable and generalizable agents.

Wilson, Brian↗

NASA NTRSDate not supplied

Evaluating a Cognitive Extension for the Licklider Transmission Protocol in a Spacecraft Emulation Testbed

In space communications, particularly when involving regions beyond cislunar space, the development of advanced networking solutions is essential to address the challenges posed by limited connectivity, substantial propagation delays, and radio signal variations. This study explores a data-driven approach to the Licklider Transmission Protocol (LTP), specifically focusing on dynamically adjusting the maximum payload size of segments. Prior research has emphasized the potential benefits of dynamically adjusting this parameter, introducing the concept of Cognitive LTP. This paper presents a software implementation of Cognitive LTP (CLTP) within an open-source Delay Tolerant Networking (DTN) framework, specifically the High-rate Delay Tolerant Networking (HDTN), and experimentally evaluates its performance under realistic space conditions. Leveraging the Cognitive Ground Testbed (CGT), developed by NASA GRC for spacecraft communication emulation, this study effectively bridges the gap between theoretical advancements and practical applications. By thoroughly analyzing CLTP’s functionality within the CGT, this research offers insights into the practical implications of adaptive networking strategies, emphasizing the importance of conducting tests in relevant environments for the maturation of space communication technologies.

Delay Tolerant Networking↗

NASA NTRS2012

Mission at Mubasi - A Simulation for Leadership Development

The United States Army is investing in simulations as a way of providing practice for leader decision making. Such simulations, grounded in lessons learned from deployment experienced leaders, place less experienced and more junior leaders in challenging situations they might soon be confronted with. And given increased demands on the Army to become more efficient, while maintaining acceptable levels of mission readiness, simulations offer a cost effective complement to live field training. So too, the design parameters of such a simulation can be made to reinforce specific behavior responses which teach leaders known theory and application of effective (and ineffective) decision making. With this in mind, the Center for Army Leadership (CAL) determined that decision-making was of critical importance. Specifically, the following aspects of decision-making were viewed as particularly important for today's Army leaders: 1) Decision dilemmas, in the form of equally appealing or equally unappealing choices, such that there is no clear "right" or "wrong" choice 2) Making decisions with incomplete or ambiguous information, and 3) Predicting and experiencing second- and third-order consequences of decisions. It is decision making in such a setting or environment that Army leaders are increasingly confronted with given the full spectrum of military operations they must be prepared for. This paper details the approach and development of this decision making simulation.

Cummings, Pau;↗

NASA NTRS1972

Degradation of learned skills: Effectiveness of practice methods on visual approach and landing skill retention

Flight control and procedural task skill degradation, and the effectiveness of retraining methods were evaluated for a simulated space vehicle approach and landing under instrument and visual flight conditions. Fifteen experienced pilots were trained and then tested after 4 months either without the benefits of practice or with static rehearsal, dynamic rehearsal or with dynamic warmup practice. Performance on both the flight control and procedure tasks degraded significantly after 4 months. The rehearsal methods effectively countered procedure task skill degradation, while dynamic rehearsal or a combination of static rehearsal and dynamic warmup practice was required for the flight control tasks. The quality of the retraining methods appeared to be primarily dependent on the efficiency of visual cue reinforcement.

Sitterley, T. E.↗

NASA NTRS1993

A proposal of an architecture for the coordination level of intelligent machines

The issue of obtaining a practical, structured, and detailed description of an architecture for the Coordination Level of Center for Intelligent Robotic Systems for Sapce Exploration (CIRSSE) Testbed Intelligent Controller is addressed. Previous theoretical and implementation works were the departure point for the discussion. The document is organized as follows: after this introductory section, section 2 summarizes the overall view of the Intelligent Machine (IM) as a control system, proposing a performance measure on which to base its design. Section 3 addresses with some detail implementation issues. An hierarchic petri-net with feedback-based learning capabilities is proposed. Finally, section 4 is an attempt to address the feedback problem. Feedback is used for two functions: error recovery and reinforcement learning of the correct translations for the petri-net transitions.

Beard, Randall↗

NASA NTRSDate not supplied

Runway Configuration Management with Offline Reinforcement Learning

Runway configuration management (RCM) is a challenging task, and it affects the efficiency of the National Airspace System (NAS) and airport surface operations significantly. Each airport, depending on the geometry, capacity, local climate patterns, etc. has multiple configurations for the runway usage for arriving and departing flights. Many factors such as the incoming/outgoing traffic load, wind direction and speed, convective weather, cloud ceiling and other environmental factors might affect the choice of a runway configuration at any point in time. However, other factors such as safety measures and regulations, noise abatement, capacity of each configuration, and preference of the air traffic controllers (ATCs) can also play a significant role in selecting the configuration. A sub-optimal selection of the runway configuration, or delay in making configuration changes might result in significant increase in taxi times for aircraft on the surface of the airport, fuel and energy use of the aircraft, and maintenance costs. It can also lead to safety concerns, such as an aircraft performing one or more go-arounds before being able to land. All these factors make RCM an extremely important and challenging decision-making process for the ATCs. The current state of practice sets the runway configuration by the ATCs based on relevant information available at the time including weather, traffic, noise abatement, safety bounds, etc. This makes the decision-making process subjective based on the accuracy of the available information and the bias in human decision making. Unfortunately, this approach yields poor results (e.g., significant delays) if the predicted outcomes are uncertain and their relative impact is not well understood. This is especially evident when the uncertainty increases the size of possible predicted outcomes (combinatorial explosion in possible scenarios) that cannot be handled by human reasoning. On the other hand, an automated approach based on machine intelligence can make use of historical data and search through all (or significant amount of) possible scenarios under uncertainty and make well-informed decisions.

Milad Memarzadeh↗

NASA NTRS2019

Trends in Human Spaceflight: Failure Tolerance, High Reliability and Correlated Failure History

In a half century of human spaceflight, NASA has continuously refined agency safety and reliability requirements in response to mission demands, critical failures, and technology development. Early spacecraft, including Mercury, Gemini and Apollo vehicles, were highly reliant on dissimilar redundancy and demonstrated test margins. Later programs, such as the reusable Space Transportation System (STS) and International Space Station (ISS), introduced probabilistic studies and isolated two-failure tolerance to improve robustness at the expense of added complexity. More recently, the Orion Multi-Program Crew Vehicle (MPCV) program adopted universal single-failure tolerance with two categorical exceptions; Zero-Failure Tolerant (0FT) and Design for Minimum Risk (DFMR) hardware. Failure tolerance variances are defined and managed in accordance with agency human-rating requirements, and require concurrence from program Technical Authorities (TA) as well as the MPCV Safety and Mission Assurance Safety and Engineering Review Panel (MSERP). To understand and reaffirm standards applied to Apollo, Space Shuttle and Orion vehicles, Orion and Deep Space Gateway Safety and Mission Assurance (S&MA) representatives conducted accelerated research to compare unique safety and reliability criteria against ground and flight anomalies, based on information contained in post-mission reports and the Problem Reporting and Corrective Action (PRACA) database. In some cases, high-profile failures and narrow escapes have reinforced decisions to maintain or adapt safety requirements. In others, empirical trends have highlighted the need for vigilance and innovative safety guidelines. Given the inability to achieve absolute compliance with evolving safety and reliability requirements, the team conducted a targeted review of DFMR and 0FT propulsion elements within the framework of changing system design, inspection, materials and process developments to formulate conclusions on technological maturity, failure density, and net changes in safety risk. Based on the aggregate performance of high-reliability and failure-tolerant systems, the authors have attempted to establish best practices and guidelines to inform future program decisions. On a somewhat cautionary note, this study is not intended to direct a universal set of requirements for future missions based on prior lessons learned. Spacecraft safety is a multi-variable problem, and attempts to mitigate past failures will not guarantee future success. However, this assessment offers a retrospective review of policy changes, implementation and effectiveness. In the future, NASA, European Space Agency (ESA) and industry partners may benefit from a more robust correlation between requirements and performance, as space-faring nations work toward more challenging, complex and long-duration commercial and deep-space ventures.

Green, Carrie↗

NASA NTRS2011

Cognitive Behavioral Training and Education for Spaceflight Operations

Cognitive behavioral-training (CBT) is an evidence-based practice commonly used to help treat insomnia, and is part of NASA's countermeasure regimen for Fatigue Management. CBT addresses the life style and habits of individuals that are maladaptive to managing stress and fatigue. This includes addressing learned behavioral responses that may cause stress and lead to an increased sense of fatigue. While the initial cause of onset of fatigue in the individual may be no longer present, the perception and engrained anticipation of fatigue persist and cause an exaggerated state of tension. CBT combined with relaxation training allows the individual to unlearn the maladaptive beliefs and behaviors and replace them with routines and techniques that allow cognitive restructuring and resultant relief from stress. CBT allows for elimination in individuals of unwanted ruminating thoughts and anticipatory anxiety by, for example, training the individuals to practice stressful situations in a relaxed state. As a result of CBT, relaxation can be accomplished in many ways, such as progressive muscle relaxation, meditation and guided imagery. CBT is not therapy, but rather the synthesis of behavioral countermeasures. CBT utilizes progressive relaxation as a means of reinforcing educational and cognitive countermeasures. These countermeasures include: masking, elimination of distracting thoughts, anxiety control, split attention, cognitive restructuring and other advanced psychological techniques.

Moonmaw, Ronald↗

NASA NTRS1995

Attitude determination using an adaptive multiple model filtering Scheme

Attitude determination has been considered as a permanent topic of active research and perhaps remaining as a forever-lasting interest for spacecraft system designers. Its role is to provide a reference for controls such as pointing the directional antennas or solar panels, stabilizing the spacecraft or maneuvering the spacecraft to a new orbit. Least Square Estimation (LSE) technique was utilized to provide attitude determination for the Nimbus 6 and G. Despite its poor performance (estimation accuracy consideration), LSE was considered as an effective and practical approach to meet the urgent need and requirement back in the 70's. One reason for this poor performance associated with the LSE scheme is the lack of dynamic filtering or 'compensation'. In other words, the scheme is based totally on the measurements and no attempts were made to model the dynamic equations of motion of the spacecraft. We propose an adaptive filtering approach which employs a bank of Kalman filters to perform robust attitude estimation. The proposed approach, whose architecture is depicted, is essentially based on the latest proof on the interactive multiple model design framework to handle the unknown of the system noise characteristics or statistics. The concept fundamentally employs a bank of Kalman filter or submodel, instead of using fixed values for the system noise statistics for each submodel (per operating condition) as the traditional multiple model approach does, we use an on-line dynamic system noise identifier to 'identify' the system noise level (statistics) and update the filter noise statistics using 'live' information from the sensor model. The advanced noise identifier, whose architecture is also shown, is implemented using an advanced system identifier. To insure the robust performance for the proposed advanced system identifier, it is also further reinforced by a learning system which is implemented (in the outer loop) using neural networks to identify other unknown quantities such as spacecraft dynamics parameters, gyro biases, dynamic disturbances, or environment variations.

Lam, Quang↗