Results for “safe reinforcement learning” — Page 2

NASA NTRS2022

SoMoGym: A Toolkit for Developing and Evaluating Controllers and Reinforcement Learning Algorithms for Soft Robots

Soft robotsoffer a host of benefits over traditional rigid robots, including inherent compliance that lets them passively adapt to variable environments and operate safely around humans and fragile objects. However, that same compliance makes it hard to use model-based methods in planning tasks requiring high precision or complex actuation sequences. Reinforcement learning (RL) can potentially find effective control policies, but training RL using physical soft robots is often infeasible, and training using simulations has had a high barrier to adoption. To accelerate research in control and RL for soft robotic systems, we introduce SoMoGym ( So ft Mo tion Gym ), a software toolkit that facilitates training and evaluating controllers for continuum robots. SoMoGym provides a set of benchmark tasks in which soft robots interact with various objects and environments. It allows evaluation of performance on these tasks for controllers of interest, and enables the use of RL to generate new controllers. Custom environments and robots can likewise be added easily. We provide and evaluate baseline RL policies for each of the benchmark tasks. These results show that SoMoGym enables the use of RL for continuum robots, a class of robots not covered by existing benchmarks, giving them the capability to autonomously solve tasks that were previously unattainable.

Moritz A. Graule↗

DOE OSTI2025

Reinforcement Learning for Anomaly Detection in Nuclear Power Plant Operation and Maintenance

In nuclear power plants (NPPs), timely identification of sensor and human errors is critical to ensure safe and efficient plant operations. Anomaly detection models can be employed for this task. However, traditional anomaly detection approaches may have high dependency on labeled datasets and struggle with adaptability in complex, dynamic environments. Reinforcement learning (RL) has demonstrated significant potential in fault diagnosis and anomaly detection; however, its application to anomaly detection in NPPs remains a relatively underexplored research direction. Hence, to address this gap, in this study, we present a novel physics-informed reinforcement learning model, PIRL-AD: Physics-Informed Reinforcement Learning for Anomaly Detection, that integrates domain knowledge from calorimetric equations into the RL framework for enhanced sensor and human error anomaly detection. We evaluate the performance of PIRL-AD against a non-physics informed RL benchmark and a support vector machine (SVM) on data collected from a forced flow loop testbed. Experimental results suggest that PIRL-AD outperforms other baselines on a range of anomalous datasets that include both sensor and human-induced anomalies across key performance metrics, statistically outperforming the RL and SVM benchmarks with respect to geometric mean (respectively, 92.96% vs. 91.06% vs. 83.01%) and F1-score (respectively, 89.23% vs. 86.98% vs. 77.01%). Furthermore, the findings suggest the potential of physics-integrated reinforcement learning models for enhanced anomaly detection performance in NPPs.

Reinforcement learning↗

DOE OSTI2025

A Sequential Model Predictive and Deep Reinforcement Learning-Based Controller for Distribution System Outage Mitigation under Hurricane Events

This paper proposes a proactive outage mitigation framework for power distribution networks to withstand hurricane-induced disruptions. It leverages Model Predictive Control (MPC) to identify safe lines for proactive switching during hurricanes, minimizing the risk of cascading failures and voltage violations. The switching strategies optimized by MPC are sequentially integrated with a Deep Reinforcement Learning agent using the Advantage Actor-Critic algorithm, enabling dynamic line switching to maximize connected buses and minimize voltage violations in real time. Using a probabilistic hurricane model, the framework predicts line failures and adapts to varying conditions to enhance grid resilience. Simulations on the IEEE 123-bus system demonstrate its effectiveness in maintaining high connectivity and minimizing disruptions. Real-time testing with an RTDS confirms the practicality and reliability of the proposed approach.

Selim, Alaa [University of Connecticut]↗

DOE OSTI2025

An analysis of physics limited dispatch of nuclear renewable integrated energy systems using deep reinforcement learning and dynamic modeling

Previous approaches to dispatching nuclear integrated energy systems (NIES) have focused on the profitability and flexibility of these systems to operate on energy grids with highly variable pricing. However, due to the complexity involved in modeling and designing these systems, there has been less emphasis on ensuring that these dispatch strategies are physically achievable. It is imperative to develop methods that allow the system to remain within the desired NIES operating conditions and perform this based on realistic limited forecasted information. This research employs next generation artificial intelligence, namely deep reinforcement learning (DRL), and a dynamic system model written in Modelica to find a safe and profitable dispatch strategy for a solar nuclear hybrid design. The DRL agent is shown to find a novel dispatch strategy that manages both power ramping and power levels while respecting operational limits. This DRL-based dispatch is compared to other dispatching strategies including an optimal design solution from mixed integer linear programming (MILP). It is found that incorporating the physics of such a tightly coupled NIES limits the profitability of the MILP-based dispatch strategy. As a result, the MILP solution overestimates the design’s generated revenue. In contrast, DRL significantly reduces the number of breaches of safe operational conditions during energy arbitrage while maintaining profitability. Furthermore, this work paves the way for a more detailed assessment of NIES profitability and could be used to aid operator decisions on future NIES projects.

14 - SOLAR ENERGY↗

NASA NTRSDate not supplied

Joint Communication Resource Allocation and Velocity Selection in Urban Air Mobility via Multi-agent Reinforcement Learning

With traffic congestion problems becoming more severe in urban areas, the National Aeronautics and Space Administration promotes the Urban Air Mobility (UAM) concept, which envisages a safe and efficient air transportation system. However, the increased communication demands in UAM can exacerbate the spectrum scarcity. Therefore, a new communication resource allocation solution is necessary. In this paper, we focus on uplink UAM communications, where multiple aerial vehicles (AV) perform cargo/passenger delivery tasks. With predefined flight paths, AVs make decisions on communication resource allocation and velocity selection to complete their missions under safety constraints. Accordingly, we formulate a joint optimization problem to minimize the weighted sum of the total travel time and communication outage time. We first model the optimization problem as a Markov game and propose a multi-agent reinforcement learning based solution. Simulation results corroborate the effectiveness of the proposed solution.

Ruixuan Han↗

DOE OSTI2020

Multi-Agent Safe Policy Learning for Power Management of Networked Microgrids

This paper presents a supervised multi-agent safe policy learning (SMAS-PL) method for optimal power management of networked microgrids (MGs) in distribution systems. While unconstrained reinforcement learning (RL) algorithms are black-box decision models that could fail to satisfy grid operational constraints, our proposed method considers AC power flow equations and other operational limits. Accordingly, the training process employs the gradient information of operational constraints to ensure that the optimal control policy functions generate safe and feasible decisions. Furthermore, we have developed a distributed consensus-based optimization approach to train the agents’ policy functions while maintaining MGs’ privacy and data ownership boundaries. After training, the learned optimal policy functions can be safely used by the MGs to dispatch their local resources, without the need to solve a complex optimization problem from scratch. Lastly, numerical experiments have been devised to verify the performance of the proposed method.

, policy gradient↗

DOE OSTI2023

Role of reinforcement learning for risk-based robust control of cyber-physical energy systems

Critical infrastructures such as cyber-physical energy systems (CPS-E) integrate information flow and physical operations that are vulnerable to natural and targeted failures. Safe, secure, and reliable operation and control of CPS-E is critical to ensure societal well-being and economic prosperity. Automated control is key for real-time operations and may be mathematically cast as a sequential decision-making problem under uncertainty. Emergence of data-driven techniques for decision making under uncertainty, such as reinforcement learning (RL), have led to promising advances for addressing sequential decision-making problems for risk-based robust CPS-E control. However, existing research challenges include understanding the applicability of RL methods across diverse CPS-E applications, addressing the effect of risk preferences across multiple RL methods, and development of open-source domain-aware simulation environments for RL experimentation within a CPS-E context. This article systematically analyzes the applicability of four types of RL methods (model-free, model-based, hybrid model-free and model-based, and hierarchical) for risk-based robust CPS-E control. Problem features and solution stability for the RL methods are also discussed. We demonstrate and compare the performance of multiple RL methods under different risk specifications (risk-averse, risk-neutral, and risk-seeking) through the development and application of an open-source simulation environment. Motivating numerical simulation examples include representative single-zone and multizone building control use cases. Finally, six key insights for future research and broader adoption of RL methods are identified, with specific emphasis on problem features, algorithmic explainability, and solution stability.

24 POWER TRANSMISSION AND DISTRIBUTION↗

NASA NTRS2020

Using Reinforcement Learning to Design Missed Thrust Resilient Trajectories

From ion thrusters to solar sails, spacecraft continue to adopt new and more efficient forms of propulsion. As these low-thrust propulsion meth- ods have become more prevalent, new challenges have arisen. Depending on the mission, low-thrust propulsion elements may need to thrust con- tinuously for days/months. During these thrusting periods, external fac- tors, such as a micro-meteoroid impact or a software glitch, may cause the spacecraft to prematurely cease its thrust stage. Half of all deep space missions enter a safe mode where they cannot thrust every four months. These missed thrust events can result in the complete loss of a space- craft for time-dependent trajectories like planetary rendezvous. This paper demonstrates how neural networks, trained using reinforcement learning, can autonomously correct for missed thrust events during an interplanetary trajectory.

Laipert, Frank E.↗

NASA NTRSDate not supplied

Dynamic Spectrum Allocation in Urban Air Transportation System via Deep Reinforcement Learning

The emerging concepts of Urban Air Mobility (UAM) and Advanced Air Mobility (AAM) open a new paradigm for urban air transportation. A big challenge is that these new aerial vehicles will quickly saturate the already crowded aviation spectrum, which is an essential resource to ensure reliable communications for safe operations. In this paper, we consider an air transportation system where multiple aerial vehicles are operated to transport passengers or cargo from different sources to destinations along their pre-defined paths. During the flight, the minimum communication Quality of Service (QoS) requirement must be achieved to ensure flight safety. Our objective is to minimize the average mission completion time by jointly optimizing the velocity selection and spectrum allocation for all aerial vehicles. We formulate the optimization problem as a multi-stage Markov Decision Process (MDP) where the optimization variables are coupled together. A multi-agent Deep Reinforcement Learning (DRL) based solution is proposed where Value Decomposition Networks (VDN) algorithm is utilized to take discrete actions. Additionally, we propose a heuristic greedy algorithm as a baseline solution. Simulation results show that our learning based solution outperforms the heuristic greedy algorithm and another Orthogonal Multiple Access (OMA) solution in minimizing the mission completion time.

Ruixuan Han↗

NASA NTRSDate not supplied

Dynamic Spectrum Allocation in Urban Air Transportation System via Deep Reinforcement Learning

The emerging concepts of Urban Air Mobility (UAM) and Advanced Air Mobility (AAM) open a new paradigm for urban air transportation. A big challenge is that these new aerial vehicles will quickly saturate the already crowded aviation spectrum, which is an essential resource to ensure reliable communications for safe operations. In this paper, we consider an air transportation system where multiple aerial vehicles are operated to transport passengers or cargo from different sources to destinations along their pre-defined paths. During the flight, the minimum communication Quality of Service (QoS) requirement must be achieved to ensure flight safety. Our objective is to minimize the average mission completion time by jointly optimizing the velocity selection and spectrum allocation for all aerial vehicles. We formulate the optimization problem as a multi-stage Markov Decision Process (MDP) where the optimization variables are coupled together. A multi-agent Deep Reinforcement Learning (DRL) based solution is proposed where Value Decomposition Networks (VDN) algorithm is utilized to take discrete actions. Additionally, we propose a heuristic greedy algorithm as a baseline solution. Simulation results show that our learning based solution outperforms the heuristic greedy algorithm and another Orthogonal Multiple Access (OMA) solution in minimizing the mission completion time.

Ruixuan Han↗

DOE OSTI2022

Moving the needle: Employing deep reinforcement learning to push the boundaries of coarse-grained vaccine models

Highly mutable infectious disease pathogens (hm-IDPs) such as HIV and influenza evolve faster than the human immune system can contain them, allowing them to circumvent traditional vaccination approaches and causing over one million deaths annually. Agent-based models can be used to simulate the complex interactions that occur between immune cells and hm-IDP-like proteins (antigens) during affinity maturation—the process by which antibodies evolve. Compared to existing experimental approaches, agent-based models offer a safe, low-cost, and rapid route to study the immune response to vaccines spanning a wide range of design variables. However, the highly stochastic nature of affinity maturation and vast sequence space of hm-IDPs render brute force searches intractable for exploring all pertinent vaccine design variables and the subset of immunization protocols encompassed therein. To address this challenge, we employed deep reinforcement learning to drive a recently developed agent-based model of affinity maturation to focus sampling on immunization protocols with greater potential to improve the chosen metrics of protection, namely the broadly neutralizing antibody (bnAb) titers or fraction of bnAbs produced. Using this approach, we were able to coarse-grain a wide range of vaccine design variables and explore the relevant design space. Our work offers new testable insights into how vaccines should be formulated to maximize protective immune responses to hm-IDPs and how they can be minimally tailored to account for major sources of heterogeneity in human immune responses and various socioeconomic factors. Our results indicate that the first 3 to 5 immunizations, depending on the metric of protection, should be specially tailored to achieve a robust protective immune response, but that beyond this point further immunizations require only subtle changes in formulation to sustain a durable bnAb response.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2023

Interpreting Primal-Dual Algorithms for Constrained Multiagent Reinforcement Learning

Constrained multiagent reinforcement learning (C-MARL) is gaining importance as MARL algorithms find new applications in real-world systems ranging from energy systems to drone swarms. Most C-MARL algorithms use a primal-dual approach to enforce constraints through a penalty function added to the reward. In this paper, we study the structural effects of this penalty term on the MARL problem. First, we show that the standard practice of using the constraint function as the penalty leads to a weak notion of safety. However, by making simple modifications to the penalty term, we can enforce meaningful probabilistic (chance and conditional value at risk) constraints. Second, we quantify the effect of the penalty term on the value function, uncovering an improved value estimation procedure. We use these insights to propose a constrained multiagent advantage actor critic (C-MAA2C) algorithm. Simulations in a simple constrained multiagent environment affirm that our reinterpretation of the primal-dual method in terms of probabilistic constraints is effective, and that our proposed value estimate accelerates convergence to a safe joint policy.

chance constraints↗

DOE OSTI2023

Approaches for Synthesis and Deployment of Controller Models on Automated Vehicles for Car-following in Mixed Autonomy

This paper describes the software design patterns and vehicle interfaces that were employed to transition vehicle controllers from simulation environments to open-road field experiments. The approach relies on a life cycle that utilizes model-based design and code generation, along with agile software development, and both software and hardware-in-the-loop testing, with additional safety margins. Autonomous designs should consider the dynamics of mixed autonomy in traffic to safely operate among humans. The software that provides a vehicle’s behavior intelligence is often developed through simulation, which may have a mismatch between dynamics, or as a result of a reinforcement learning workflow, which may be a black box with challenges to analyze. In each of these cases, it is important to have research interfaces that provide strongly typed data streams accessible to researchers who are not software experts while continuing to satisfy safety and liveness constraints. This paper describes how we design the hardware platform interfaces and software design process for a mixed autonomy traffic experiment with a leader-follower scenario. Controller synthesis for these vehicles requires clearly articulated vehicle interfaces and software design patterns for successful onboard deployment. Testing strategies for such controllers are also described before algorithms are transitioned to full-scale field experiments with safety operators for the vehicles. Testing strategies include software-in-the-loop simulation testing, hardware-in-the-loop simulation, ghost-car testing, and read-only testing in live traffic. With our approach, we were not only able to validate our controller synthesized in scripts and simulation, but also able to scale deployment to multiple vehicles.

Bhadani, Rahul↗

NASA NTRS2018

Initial Approach to Collect Small Unmanned Aircraft System Off-Nominal Operational Situations Data

NASA is developing the Unmanned Aircraft System Traffic Management research platform to safely integrate small unmanned aircraft operations in large-scale at low-altitudes. As a part of this effort, small unmanned aircraft system off-nominal operational situations data collection process has been developed to take lessons learned and to reinforce operational compliance. In this paper, descriptions of variables used for digital data collection and an online report form for collection of observational data from the operators (contextual data) are provided. They are used to collect off-nominal data from the Unmanned Aircraft System Traffic Management National Campaign in 2017. The digital data show that 2 out of 118 campaign operations (1.7%) encountered loss of navigation. Since the campaign aircraft used Global Positioning System for navigation, it is likely that unobstructed view of the sky at the campaign locations contributed to this small number. Also, 4 out of 47 operations (8.5%) encountered loss of communications. A relatively short distance between ground control system and aircraft, ranging from 2300 feet to 4200 feet, likely contributed to this small number. There was no data to identify the loss of communications condition, aircraft received signal strength, for the remaining 71 operations suggesting that some operators may not be monitoring unmanned aircraft communications system performance or monitoring it with different parameters. For the contextual data, due to the low number of total reports during the campaign, no significant trends emerged. This is an initial attempt to collect contextual data from small unmanned aircraft operators about off-nominal situations, and changes will be made to the future data collection to improve the amount and quality of the information.

unmanned aviation systems traffic management (UTM)↗

NASA NTRS2018

Initial Approach to Collect Small Unmanned Aircraft System Off-Nominal Operational Situations Data

NASA is developing the Unmanned Aircraft System Traffic Management research platform to safely integrate small unmanned aircraft operations in large-scale at low-altitudes. As a part of this effort, small unmanned aircraft system off-nominal operational situations data collection process has been developed to take lessons learned and to reinforce operational compliance. In this paper, descriptions of variables used for digital data collection and an online report form for collection of observational data from the operators (contextual data) are provided. They are used to collect off-nominal data from the Unmanned Aircraft System Traffic Management National Campaign in 2017. The digital data show that 2 out of 118 campaign operations (1.7%) encountered loss of navigation. Since the campaign aircraft used Global Positioning System for navigation, it is likely that unobstructed view of the sky at the campaign locations contributed to this small number. Also, 4 out of 47 operations (8.5%) encountered loss of communications. A relatively short distance between ground control system and aircraft, ranging from 2300 feet to 4200 feet, likely contributed to this small number. There was no data to identify the loss of communications condition, aircraft received signal strength, for the remaining 71 operations suggesting that some operators may not be monitoring unmanned aircraft communications system performance or monitoring it with different parameters. For the contextual data, due to the low number of total reports during the campaign, no significant trends emerged. This is an initial attempt to collect contextual data from small unmanned aircraft operators about off-nominal situations, and changes will be made to the future data collection to improve the amount and quality of the information.

Jung, Jaewoo↗

DOE OSTI2024

Development of algorithms for augmenting and replacing conventional process control using reinforcement learning

Here, this work seeks to allow for the online operation and training of model-free reinforcement learning (RL) agents but limit the risk to system equipment and personnel. The parallel implementation of RL alongside more conventional process control (CPC) allows for the RL algorithm to learn from CPC. The past performance of both methods are assessed on a continuous basis allowing for a transition from CPC to RL and, if needed, transitioning back to CPC from RL. This allows for the RL algorithm to slowly and safely assume control of the process without significant degradation in control performance. It is shown that the RL can derive a near optimal policy even when coupled with a suboptimal CPC. It is also demonstrated that the coupled RL-CPC algorithm learns at a faster rate than traditional RL methods of exploration while the algorithm’s performance does not deteriorate below CPC, even when exposed to an unknown operating condition.

30 DIRECT ENERGY CONVERSION↗

DOE OSTI2021

Barrier Function-based Reinforcement Learning for Emergency Control of Power Systems

Under voltage load shedding has been considered as a standard and effective measure to recover the voltage stability of the electric power grid under emergency and severe conditions. However, this scheme usually trips a massive amount of load which can be unnecessary and harmful to customers. Recently, deep reinforcement learning (RL) has been regarded and adopted as a promising approach that can significantly reduce the amount of load shedding. However, like most existing machine learning (ML)-based control techniques, RL control usually cannot guarantee the safety of the systems under control. In this paper, we introduce a novel safe RL method for load shedding of power systems, that can enhance the safe voltage recovery of the electric power grid after experiencing faults. Unlike standard RL method, the safe RL method integrates a Barrier function into the reward function. Consequently, the optimal control policy can render the power system to avoid the safety bounds. This method is general and can be applied to other safety-critical control problems. Numerical simulations on the 39-bus IEEE benchmark is performed to demonstrate the effectiveness of the proposed safe RL emergency control, as well as its adaptive capability to faults not seen in the training.

Vu, Thanh Long↗

DOE OSTI2024

Machine Learning-Driven Quantification of CO2 Plume Dynamics at Illinois Basin Decatur Project Sites Using Microseismic Data

This study utilizes machine learning to quantify CO2 plume extents by analyzing microseismic data from the Illinois Basin Decatur Project (IBDP). Leveraging a unique dataset of well logs, microseismic records, and CO2 injection metrics, this work aims to predict the temporal evolution of subsurface CO2 saturation plumes. The findings illustrate that machine learning can predict plume dynamics, revealing vertical clustering of microseismic events over distinct time periods within certain proximities to the injection well, consistent with an invasion percolation model. The buoyant CO2 plume partially trapped within sandstone intervals periodically breaches localized barriers or baffles, which act as leaky seals and impede vertical migration until buoyancy overcomes gravity and capillary forces, leading to breakthroughs along vertical zones of weakness. Between different unsupervised clustering techniques, K-Means and DBSCAN were applied and analyzed in detail, where K-means outperformed DBSCAN in this specific study by indicating the combination of the highest Silhouette Score and the lowest Davies–Bouldin Index. The predictive capability of machine learning models in quantifying CO2 saturation plume extension is significant for real-time monitoring and management of CO2 sequestration sites. The models exhibit high accuracy, validated against physical models and injection data from the IBDP, reinforcing the viability of CO2 geological sequestration as a climate change mitigation strategy and enhancing advanced tools for safe management of these operations.

Iyegbekedo, Ikponmwosa↗