Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “multi-agent reinforcement learning”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

A Novel LDPP-MADDPG Approach for Distributed Power Allocation in mmWave Cellular Networks

This paper considers the problem of distributed beam scheduling and power allocation problem in millimeter- Wave (mmWave) cellular networks, in which multiple Base Stations (BSs) operate as individual operators over a shared spectrum. We propose a novel learning-aided approach that integrates the Lyapunov Drift-Plus-Penalty (LDPP) framework and Multi-agent Deep Deterministic Policy Gradient (MADDPG) reinforcement learning algorithms. This offers a powerful approach to learning stable and constraint-aware policies, reaping the joint benefit of both LDPP and MADDPG, in complex multiagent environments. The major challenge for this approach is to integrate these two approaches in a meaningful and effective manner. The key idea to solve this problem is to introduce a novel feature of local observation that incorporates potential negative value of the reward function due to the stochastic constraints introduced by the LDPP framework. Empirical results demonstrate that our proposed scheme outperforms the baseline methods under various conditions.

99 - GENERAL AND MISCELLANEOUS↗
NASA NTRS2001

Learning Sequences of Actions in Collectives of Autonomous Agents

In this paper we focus on the problem of designing a collective of autonomous agents that individually learn sequences of actions such that the resultant sequence of joint actions achieves a predetermined global objective. We are particularly interested in instances of this problem where centralized control is either impossible or impractical. For single agent systems in similar domains, machine learning methods (e.g., reinforcement learners) have been successfully used. However, applying such solutions directly to multi-agent systems often proves problematic, as agents may work at cross-purposes, or have difficulty in evaluating their contribution to achievement of the global objective, or both. Accordingly, the crucial design step in multiagent systems centers on determining the private objectives of each agent so that as the agents strive for those objectives, the system reaches a good global solution. In this work we consider a version of this problem involving multiple autonomous agents in a grid world. We use concepts from collective intelligence to design goals for the agents that are 'aligned' with the global goal, and are 'learnable' in that agents can readily see how their behavior affects their utility. We show that reinforcement learning agents using those goals outperform both 'natural' extensions of single agent algorithms and global reinforcement, learning solutions based on 'team games'.

Turner, Kagan↗
NASA NTRS2004

Time-Extended Policies in Mult-Agent Reinforcement Learning

Reinforcement learning methods perform well in many domains where a single agent needs to take a sequence of actions to perform a task. These methods use sequences of single-time-step rewards to create a policy that tries to maximize a time-extended utility, which is a (possibly discounted) sum of these rewards. In this paper we build on our previous work showing how these methods can be extended to a multi-agent environment where each agent creates its own policy that works towards maximizing a time-extended global utility over all agents actions. We show improved methods for creating time-extended utilities for the agents that are both "aligned" with the global utility and "learnable." We then show how to crate single-time-step rewards while avoiding the pi fall of having rewards aligned with the global reward leading to utilities not aligned with the global utility. Finally, we apply these reward functions to the multi-agent Gridworld problem. We explicitly quantify a utility's learnability and alignment, and show that reinforcement learning agents using the prescribed reward functions successfully tradeoff learnability and alignment. As a result they outperform both global (e.g., team games ) and local (e.g., "perfectly learnable" ) reinforcement learning solutions by as much as an order of magnitude.

Tumer, Kagan↗
NASA NTRS2004

Multi-Agent Methods for the Configuration of Random Nanocomputers

As computational devices continue to shrink, the cost of manufacturing such devices is expected to grow exponentially. One alternative to the costly, detailed design and assembly of conventional computers is to place the nano-electronic components randomly on a chip. The price for such a trivial assembly process is that the resulting chip would not be programmable by conventional means. In this work, we show that such random nanocomputers can be adaptively programmed using multi-agent methods. This is accomplished through the optimization of an associated high dimensional error function. By representing each of the independent variables as a reinforcement learning agent, we are able to achieve convergence must faster than with other methods, including simulated annealing. Standard combinational logic circuits such as adders and multipliers are implemented in a straightforward manner. In addition, we show that the intrinsic flexibility of these adaptive methods allows the random computers to be reconfigured easily, making them reusable. Recovery from faults is also demonstrated.

Lawson, John W.↗
DOE OSTI2026

Wolf

The Workflow Orchestration Language Framework (WOLF) is an agentic framework grounded in natural language with an architecture inspired by reinforcement learning (RL)—designed to orchestrate, scale, and accelerate complex workflows. The concept of WOLF was born out of the very successful ASC Tri-lab Multi-Agent Design Assistant (MADA) project, but extends beyond its domain-specific design agents to provide a more general and extensible architecture. WOLF capitalizes on the lessons learned from MADA and is fully aligned with Sutton’s The Bitter Lesson—that the most enduring progress in AI comes from general-purpose methods that scale with computation, rather than narrow techniques built on domain-specific human knowledge. In this spirit, WOLF enables agents to autonomously learn workflows, capture strategies as reusable playbooks, and build a growing corpus of interpretable, auditable “wisdom artifacts.” These artifacts, expressed in natural language, bridge human and machine understanding while preserving adaptability and scalability as computational power continues to expand.

Boureima, Ismaeal↗
DOE OSTI2025

CityLearn v2: energy-flexible, resilient, occupant-centric, and carbon-aware management of grid-interactive communities

As more distributed energy resources become part of the demand-side infrastructure, quantifying their energy flexibility on a community scale is crucial. CityLearn v1 provided an environment for benchmarking control algorithms. However, there is no standardized environment utilizing realistic building-stock datasets for distributed energy resource control benchmarking without co-simulation or third-party frameworks. CityLearn v2 extends CityLearn v1 by providing a stand-alone simulation environment that leverages the End-Use Load Profiles for the U.S. Building Stock dataset to create grid-interactive communities for resilient, multi-agent, and objective control of distributed energy resources with dynamic occupant feedback. While the v1 environment used pre-simulated building thermal loads, the v2 environment uses data-driven thermal dynamics and eliminates the need for co-simulation with building energy performance software. This work details the v2 environment and provides application examples that use reinforcement learning control to manage battery energy storage system, vehicle-to-grid control, and thermal comfort during heat pump power modulation.

Nweye, Kingsley↗
NASA NTRS2004

Coordination in Large Collectives

Finding the subset of a set of imperfect devices (e.g., nano or micro devices) that results in the best aggregate device is a challenging problem. It is an abstraction of what will likely be a major difficulty in designing and controlling systems of nano or micro-scale components, particularly when a large fraction of those components may be unreliable. Rather than approaching this as a passive search problem, we transform the problem into one of coordination in a complex system by imbuing each device with simple decision making ability. In doing so we face the challenge of determining what each component should attempt to do so that the collective behavior solves the overall problem. Furthermore, in this padicular instance, we face problems of scaling (number of components in the thousands to tens of thousands), observability (components have limited sensing capabilities), and reliability (the components are faulty). We present an approach based on deriving component goals that are aligned with the overall system goal (e.g., forming best aggregate device), and can be computed using information readily (e.g., locally) available to the components. Then, each component in such a collective uses a simple reinforcement learning algorithm to selfishly pursue its own goals. Because those goals are derived in a principled manner, there is no need to use external mechanisms to force collaboration or coordination among the components to ensure that the system reaches a globally desirable solution. The results show that not only this approach provides improvements of over an order of magnitude over both traditional search methods and traditional multi-agent methods, but that the gains increase with the size of the system. This latter result makes this method ideal for domains where the number of components is currently in the thousands and will reach millions in the near future.

Tumer, Kagan↗
NASA NTRS2004

Unifying Temporal and Structural Credit Assignment Problems

Single-agent reinforcement learners in time-extended domains and multi-agent systems share a common dilemma known as the credit assignment problem. Multi-agent systems have the structural credit assignment problem of determining the contributions of a particular agent to a common task. Instead, time-extended single-agent systems have the temporal credit assignment problem of determining the contribution of a particular action to the quality of the full sequence of actions. Traditionally these two problems are considered different and are handled in separate ways. In this article we show how these two forms of the credit assignment problem are equivalent. In this unified frame-work, a single-agent Markov decision process can be broken down into a single-time-step multi-agent process. Furthermore we show that Monte-Carlo estimation or Q-learning (depending on whether the values of resulting actions in the episode are known at the time of learning) are equivalent to different agent utility functions in a multi-agent system. This equivalence shows how an often neglected issue in multi-agent systems is equivalent to a well-known deficiency in multi-time-step learning and lays the basis for solving time-extended multi-agent problems, where both credit assignment problems are present.

Agogino, Adrian K.↗
NASA NTRS2005

Designing Agent Utilities for Coordinated, Scalable and Robust Multi-Agent Systems

Coordinating the behavior of a large number of agents to achieve a system level goal poses unique design challenges. In particular, problems of scaling (number of agents in the thousands to tens of thousands), observability (agents have limited sensing capabilities), and robustness (the agents are unreliable) make it impossible to simply apply methods developed for small multi-agent systems composed of reliable agents. To address these problems, we present an approach based on deriving agent goals that are aligned with the overall system goal, and can be computed using information readily available to the agents. Then, each agent uses a simple reinforcement learning algorithm to pursue its own goals. Because of the way in which those goals are derived, there is no need to use difficult to scale external mechanisms to force collaboration or coordination among the agents, or to ensure that agents actively attempt to appropriate the tasks of agents that suffered failures. To present these results in a concrete setting, we focus on the problem of finding the sub-set of a set of imperfect devices that results in the best aggregate device. This is a large distributed agent coordination problem where each agent (e.g., device) needs to determine whether to be part of the aggregate device. Our results show that the approach proposed in this work provides improvements of over an order of magnitude over both traditional search methods and traditional multi-agent methods. Furthermore, the results show that even in extreme cases of agent failures (i.e., half the agents failed midway through the simulation) the system's performance degrades gracefully and still outperforms a failure-free and centralized search algorithm. The results also show that the gains increase as the size of the system (e.g., number of agents) increases. This latter result is particularly encouraging and suggests that this method is ideally suited for domains where the number of agents is currently in the thousands and will reach tens or hundreds of thousands in the near future.

Tumer, Kagan↗
DOE OSTI2024

ReLIC: Full-Scale Realization of Reinforcement Learning for Infrastructure Control

Prior efforts have shown that deep reinforcement learning (DRL) may provide a new method for controlling networked power systems. Though successful, prior approaches have not yet demonstrated their behavior on systems of realistic scale. This effort examined multiple theoretical and technical approaches to allow a DRL model to operate over a system of 2,000 buses or more. We find that allowing the DRL models to run training episodes in parallel provides near limitless efficiency gains, allowing us to train successful agents to behave on our Kuramoto transmission model of up to 4,000 buses. We further show that we can expand our PowerWorld DRL implementation to systems of up to 25 buses but struggle to go beyond this limit due to PowerWorld’s inability to run multiple instances at once. Finally, we examine a multi-agent approach and find that it performs as well if not better than our existing centralized approach.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2000

Adaptivity in Agent-Based Routing for Data Networks

Adaptivity, both of the individual agents and of the interaction structure among the agents, seems indispensable for scaling up multi-agent systems (MAS s) in noisy environments. One important consideration in designing adaptive agents is choosing their action spaces to be as amenable as possible to machine learning techniques, especially to reinforcement learning (RL) techniques. One important way to have the interaction structure connecting agents itself be adaptive is to have the intentions and/or actions of the agents be in the input spaces of the other agents, much as in Stackelberg games. We consider both kinds of adaptivity in the design of a MAS to control network packet routing. We demonstrate on the OPNET event-driven network simulator the perhaps surprising fact that simply changing the action space of the agents to be better suited to RL can result in very large improvements in their potential performance: at their best settings, our learning-amenable router agents achieve throughputs up to three and one half times better than that of the standard Bellman-Ford routing algorithm, even when the Bellman-Ford protocol traffic is maintained. We then demonstrate that much of that potential improvement can be realized by having the agents learn their settings when the agent interaction structure is itself adaptive.

Wolpert, David H.↗
NASA NTRS1999

Cooperation and Coordination Between Fuzzy Reinforcement Learning Agents in Continuous State Partially Observable Markov Decision Processes

Successful operations of future multi-agent intelligent systems require efficient cooperation schemes between agents sharing learning experiences. We consider a pseudo-realistic world in which one or more opportunities appear and disappear in random locations. Agents use fuzzy reinforcement learning to learn which opportunities are most worthy of pursuing based on their promise rewards, expected lifetimes, path lengths and expected path costs. We show that this world is partially observable because the history of an agent influences the distribution of its future states. We consider a cooperation mechanism in which agents share experience by using and-updating one joint behavior policy. We also implement a coordination mechanism for allocating opportunities to different agents in the same world. Our results demonstrate that K cooperative agents each learning in a separate world over N time steps outperform K independent agents each learning in a separate world over K*N time steps, with this result becoming more pronounced as the degree of partial observability in the environment increases. We also show that cooperation between agents learning in the same world decreases performance with respect to independent agents. Since cooperation reduces diversity between agents, we conclude that diversity is a key parameter in the trade off between maximizing utility from cooperation when diversity is low and maximizing utility from competitive coordination when diversity is high.

Berenji, Hamid R.↗
NASA NTRS1999

A Survey of Collective Intelligence

This chapter presents the science of "COllective INtelligence" (COIN). A COIN is a large multi-agent systems where: i) the agents each run reinforcement learning (RL) algorithms; ii) there is little to no centralized communication or control; iii) there is a provided world utility function that, rates the possible histories of tile full system. Tile conventional approach to designing large distributed systems to optimize a world utility does not use agents running RL algorithms. Rather that approach begins with explicit modeling of the overall system's dynamics, followed by detailed hand-tuning of the interactions between the components to ensure that they "cooperate" as far as the world utility is concerned. This approach is labor-intensive, often results in highly non-robust systems, and usually results in design techniques that, have limited applicability. In contrast, with COINs we wish to solve the system design problems implicitly, via the 'adaptive' character of the RL algorithms of each of the agents. This COIN approach introduces an entirely new, profound design problem: Assuming the RL algorithms are able to achieve high rewards, what reward functions for the individual agents will, when pursued by those agents, result in high world utility? In other words, what reward functions will best ensure that we do not have phenomena like the tragedy of the commons, or Braess's paradox? Although still very young, the science of COINs has already resulted in successes in artificial domains, in particular in packet-routing, the leader-follower problem, and in variants of Arthur's "El Farol bar problem". It is expected that as it matures not only will COIN science expand greatly the range of tasks addressable by human engineers, but it will also provide much insight into already established scientific fields, such as economics, game theory, or population biology.

Wolpert, David H.↗
NASA NTRS2000

Optimal Wonderful Life Utility Functions in Multi-Agent Systems

The mathematics of Collective Intelligence (COINs) is concerned with the design of multi-agent systems so as to optimize an overall global utility function when those systems lack centralized communication and control. Typically in COINs each agent runs a distinct Reinforcement Learning (RL) algorithm, so that much of the design problem reduces to how best to initialize/update each agent's private utility function, as far as the ensuing value of the global utility is concerned. Traditional team game solutions to this problem assign to each agent the global utility as its private utility function. In previous work we used the COIN framework to derive the alternative Wonderful Life Utility (WLU), and experimentally established that having the agents use it induces global utility performance up to orders of magnitude superior to that induced by use of the team game utility. The WLU has a free parameter (the clamping parameter) which we simply set to zero in that previous work. Here we derive the optimal value of the clamping parameter, and demonstrate experimentally that using that optimal value can result in significantly improved performance over that of clamping to zero, over and above the improvement beyond traditional approaches.

Wolpert, David H.↗
NASA NTRS2001

Design and Control of Large Collections of Learning Agents

The intelligent control of multiple autonomous agents is an important yet difficult task. Previous methods used to address this problem have proved to be either too brittle, too hard to use, or not scalable to large systems. The 'Collective Intelligence' project at NASA/Ames provides an elegant, machine-learning approach to address these problems. This approach mathematically defines some essential properties that a reward system should have to promote coordinated behavior among reinforcement learners. This work has focused on creating additional key properties and algorithms within the mathematics of the Collective Intelligence framework. One of the additions will allow agents to learn more quickly, in a more coordinated manner. The other will let agents learn with less knowledge of their environment. These additions will allow the framework to be applied more easily, to a much larger domain of multi-agent problems.

Agogino, Adrian↗
NASA NTRS2004

Adaptive, Distributed Control of Constrained Multi-Agent Systems

Product Distribution (PO) theory was recently developed as a broad framework for analyzing and optimizing distributed systems. Here we demonstrate its use for adaptive distributed control of Multi-Agent Systems (MASS), i.e., for distributed stochastic optimization using MAS s. First we review one motivation of PD theory, as the information-theoretic extension of conventional full-rationality game theory to the case of bounded rational agents. In this extension the equilibrium of the game is the optimizer of a Lagrangian of the (Probability dist&&on on the joint state of the agents. When the game in question is a team game with constraints, that equilibrium optimizes the expected value of the team game utility, subject to those constraints. One common way to find that equilibrium is to have each agent run a Reinforcement Learning (E) algorithm. PD theory reveals this to be a particular type of search algorithm for minimizing the Lagrangian. Typically that algorithm i s quite inefficient. A more principled alternative is to use a variant of Newton's method to minimize the Lagrangian. Here we compare this alternative to RL-based search in three sets of computer experiments. These are the N Queen s problem and bin-packing problem from the optimization literature, and the Bar problem from the distributed RL literature. Our results confirm that the PD-theory-based approach outperforms the RL-based scheme in all three domains.

Bieniawski, Stefan↗
NASA NTRS2004

Product Distribution Theory for Control of Multi-Agent Systems

Product Distribution (PD) theory is a new framework for controlling Multi-Agent Systems (MAS's). First we review one motivation of PD theory, as the information-theoretic extension of conventional full-rationality game theory to the case of bounded rational agents. In this extension the equilibrium of the game is the optimizer of a Lagrangian of the (probability distribution of) the joint stare of the agents. Accordingly we can consider a team game in which the shared utility is a performance measure of the behavior of the MAS. For such a scenario the game is at equilibrium - the Lagrangian is optimized - when the joint distribution of the agents optimizes the system's expected performance. One common way to find that equilibrium is to have each agent run a reinforcement learning algorithm. Here we investigate the alternative of exploiting PD theory to run gradient descent on the Lagrangian. We present computer experiments validating some of the predictions of PD theory for how best to do that gradient descent. We also demonstrate how PD theory can improve performance even when we are not allowed to rerun the MAS from different initial conditions, a requirement implicit in some previous work.

Lee, Chia Fan↗
NASA NTRS2006

Agent Reward Shaping for Alleviating Traffic Congestion

Traffic congestion problems provide a unique environment to study how multi-agent systems promote desired system level behavior. What is particularly interesting in this class of problems is that no individual action is intrinsically "bad" for the system but that combinations of actions among agents lead to undesirable outcomes, As a consequence, agents need to learn how to coordinate their actions with those of other agents, rather than learn a particular set of "good" actions. This problem is ubiquitous in various traffic problems, including selecting departure times for commuters, routes for airlines, and paths for data routers. In this paper we present a multi-agent approach to two traffic problems, where far each driver, an agent selects the most suitable action using reinforcement learning. The agent rewards are based on concepts from collectives and aim to provide the agents with rewards that are both easy to learn and that if learned, lead to good system level behavior. In the first problem, we study how agents learn the best departure times of drivers in a daily commuting environment and how following those departure times alleviates congestion. In the second problem, we study how agents learn to select desirable routes to improve traffic flow and minimize delays for. all drivers.. In both sets of experiments,. agents using collective-based rewards produced near optimal performance (93-96% of optimal) whereas agents using system rewards (63-68%) barely outperformed random action selection (62-64%) and agents using local rewards (48-72%) performed worse than random in some instances.

Tumer, Kagan↗