Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “single agent”

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 19 records
All sourcesNASA NTRSDOE OSTI
NASA NTRS1991

Single-agent parallel window search

Parallel window search is applied to single-agent problems by having different processes simultaneously perform iterations of Iterative-Deepening-A(asterisk) (IDA-asterisk) on the same problem but with different cost thresholds. This approach is limited by the time to perform the goal iteration. To overcome this disadvantage, the authors consider node ordering. They discuss how global node ordering by minimum h among nodes with equal f = g + h values can reduce the time complexity of serial IDA-asterisk by reducing the time to perform the iterations prior to the goal iteration. Finally, the two ideas of parallel window search and node ordering are combined to eliminate the weaknesses of each approach while retaining the strengths. The resulting approach, called simply parallel window search, can be used to find a near-optimal solution quickly, improve the solution until it is optimal, and then finally guarantee optimality, depending on the amount of time available.

Powley, Curt↗
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 NTRS2004

Efficient Evaluation Functions for Multi-Rover Systems

Evolutionary computation can be a powerful tool in cresting a control policy for a single agent receiving local continuous input. This paper extends single-agent evolutionary computation to multi-agent systems, where a collection of agents strives to maximize a global fitness evaluation function that rates the performance of the entire system. This problem is solved in a distributed manner, where each agent evolves its own population of neural networks that are used as the control policies for the agent. Each agent evolves its population using its own agent-specific fitness evaluation function. We propose to create these agent-specific evaluation functions using the theory of collectives to avoid the coordination problem where each agent evolves a population that maximizes its own fitness function, yet the system has a whole achieves low values of the global fitness function. Instead we will ensure that each fitness evaluation function is both "aligned" with the global evaluation function and is "learnable," i.e., the agents can readily see how their behavior affects their evaluation function. We then show how these agent-specific evaluation functions outperform global evaluation methods by up to 600% in a domain where a set of rovers attempt to maximize the amount of information observed while navigating through a simulated environment.

Agogino, Adrian↗
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 NTRS2005

Distributed Evaluation Functions for Fault Tolerant Multi-Rover Systems

The ability to evolve fault tolerant control strategies for large collections of agents is critical to the successful application of evolutionary strategies to domains where failures are common. Furthermore, while evolutionary algorithms have been highly successful in discovering single-agent control strategies, extending such algorithms to multiagent domains has proven to be difficult. In this paper we present a method for shaping evaluation functions for agents that provide control strategies that both are tolerant to different types of failures and lead to coordinated behavior in a multi-agent setting. This method neither relies of a centralized strategy (susceptible to single point of failures) nor a distributed strategy where each agent uses a system wide evaluation function (severe credit assignment problem). In a multi-rover problem, we show that agents using our agent-specific evaluation perform up to 500% better than agents using the system evaluation. In addition we show that agents are still able to maintain a high level of performance when up to 60% of the agents fail due to actuator, communication or controller faults.

Agogino, Adrian↗
NASA NTRS2002

Time-Extended Payoffs for Collectives of Autonomous Agents

A collective is a set of self-interested agents which try to maximize their own utilities, along with a a well-defined, time-extended world utility function which rates the performance of the entire system. In this paper, we use theory of collectives to design time-extended payoff utilities for agents that are both aligned with the world utility, and are "learnable", i.e., the agents can readily see how their behavior affects their utility. We show that in systems where each agent aims to optimize such payoff functions, coordination arises as a byproduct of the agents selfishly pursuing their own goals. A game theoretic analysis shows that such payoff functions have the net effect of aligning the Nash equilibrium, Pareto optimal solution and world utility optimum, thus eliminating undesirable behavior such as agents working at cross-purposes. We then apply collective-based payoff functions to the token collection in a gridworld problem where agents need to optimize the aggregate value of tokens collected across an episode of finite duration (i.e., an abstracted version of rovers on Mars collecting scientifically interesting rock samples, subject to power limitations). We show that, regardless of the initial token distribution, reinforcement learning agents using collective-based payoff functions significantly outperform both natural extensions of single agent algorithms and global reinforcement learning solutions based on "team games".

Tumer, Kagan↗
NASA NTRS1999

Neuroprotective "agents" in surgery. Secret "agent" man, or common "agent" machine?

The search for clinically-effective neuroprotective agents has received enormous support in recent years--an estimated $200 million by pharmaceutical companies on clinical trials for traumatic brain injury alone. At the same time, the pathophysiology of brain injury has proved increasingly complex, rendering the likelihood of a single agent "magic bullet" even more remote. On the other hand, great progress continues with technology that makes surgery less invasive and less risky. One example is the application of endovascular techniques to treat coronary artery stenosis, where both the invasiveness of sternotomy and the significant neurological complication rate (due to microemboli showering the cerebral vasculature) can be eliminated. In this paper we review aspects of intraoperative neuroprotection both present and future. Explanations for the slow progress on pharmacologic neuroprotection during surgery are presented. Examples of technical advances that have had great impact on neuroprotection during surgery are given both from coronary artery stenosis surgery and from surgery for Parkinson's disease. To date, the progress in neuroprotection resulting from such technical advances is an order of magnitude greater than that resulting from pharmacologic agents used during surgery. The progress over the last 20 years in guidance during surgery (CT and MRI image-guidance) and in surgical access (endoscopic and endovascular techniques) will soon be complemented by advances in our ability to evaluate biological tissue intraoperatively in real-time. As an example of such technology, the NASA Smart Probe project is considered. In the long run (i.e., in 10 years or more), pharmacologic "agents" aimed at the complex pathophysiology of nervous system injury in man will be the key to true intraoperative neuroprotection. In the near term, however, it is more likely that mundane "agents" based on computers, microsensors, and microeffectors will be the major impetus to improved intraoperative neuroprotection.

Review Literature↗
NASA NTRS2024

Universal Coating by 3D Hybrid Programmable Matter

Motivated by the prospect of nano-robots that assist human physiological functions at the nanoscale, we investigate the coating problem in the three-dimensional model for hybrid programmable matter. In this model, a single agent with strictly limited viewing range and the computational capability of a deterministic finite automaton can act on passive tiles by picking up a tile, moving, and placing it at some spot. The goal of the coating problem is to fill each node of some surface graph of size n with a tile. We first solve the problem on a restricted class of graphs with a single tile type, and then use constantly many tile types to encode this graph in certain surface graphs capturing the surface of 3D objects. Our algorithm requires O(n^2) steps, which is worst-case optimal compared to an agent with global knowledge and no memory restrictions.

programmable matter↗
NASA NTRS2004

An Architecture for Controlling Multiple Robots

The Control Architecture for Multirobot Outpost (CAMPOUT) is a distributed-control architecture for coordinating the activities of multiple robots. In the CAMPOUT, multiple-agent activities and sensor-based controls are derived as group compositions and involve coordination of more basic controllers denoted, for present purposes, as behaviors. The CAMPOUT provides basic mechanistic concepts for representation and execution of distributed group activities. One considers a network of nodes that comprise behaviors (self-contained controllers) augmented with hyper-links, which are used to exchange information between the nodes to achieve coordinated activities. Group behavior is guided by a scripted plan, which encodes a conditional sequence of single-agent activities. Thus, higher-level functionality is composed by coordination of more basic behaviors under the downward task decomposition of a multi-agent planner

Aghazarian, Hrand↗
NASA NTRS1991

NASA TileWorld manual (system version 2.2)

The commands are documented of the NASA TileWorld simulator, as well as providing information about how to run it and extend it. The simulator, implemented in Common Lisp with Common Windows, encodes a particular range in a spectrum of domains, for controllable research experiments. TileWorld consists of a two dimensional grid of cells, a set of polygonal tiles, and a single agent which can grasp and move tiles. In addition to agent executable actions, there is an external event over which the agent has not control; this event correspond to a 'gust of wind'.

Philips, Andrew B.↗
NASA NTRS1992

Experiments with a decision-theoretic scheduler

This paper describes DTS, a decision-theoretic scheduler designed to employ state-of-the-art probabilistic inference technology to speed the search for efficient solutions to constraint-satisfaction problems. Our approach involves assessing the performance of heuristic control strategies that are normally hard-coded into scheduling systems, and using probabilistic inference to aggregate this information in light of features of a given problem. BPS, the Bayesian Problem-Solver, introduced a similar approach to solving single-agent and adversarial graph search problems, yielding orders-of-magnitude improvement over traditional techniques. Initial efforts suggest that similar improvements will be realizable when applied to typical constraint-satisfaction scheduling problems.

Hansson, Othar↗
NASA NTRS1993

Decision-theoretic control of EUVE telescope scheduling

This paper describes a decision theoretic scheduler (DTS) designed to employ state-of-the-art probabilistic inference technology to speed the search for efficient solutions to constraint-satisfaction problems. Our approach involves assessing the performance of heuristic control strategies that are normally hard-coded into scheduling systems and using probabilistic inference to aggregate this information in light of the features of a given problem. The Bayesian Problem-Solver (BPS) introduced a similar approach to solving single agent and adversarial graph search patterns yielding orders-of-magnitude improvement over traditional techniques. Initial efforts suggest that similar improvements will be realizable when applied to typical constraint-satisfaction scheduling problems.

Hansson, Othar↗
NASA NTRS1994

DTS: Building custom, intelligent schedulers

DTS is a decision-theoretic scheduler, built on top of a flexible toolkit -- this paper focuses on how the toolkit might be reused in future NASA mission schedulers. The toolkit includes a user-customizable scheduling interface, and a 'Just-For-You' optimization engine. The customizable interface is built on two metaphors: objects and dynamic graphs. Objects help to structure problem specifications and related data, while dynamic graphs simplify the specification of graphical schedule editors (such as Gantt charts). The interface can be used with any 'back-end' scheduler, through dynamically-loaded code, interprocess communication, or a shared database. The 'Just-For-You' optimization engine includes user-specific utility functions, automatically compiled heuristic evaluations, and a postprocessing facility for enforcing scheduling policies. The optimization engine is based on BPS, the Bayesian Problem-Solver (1,2), which introduced a similar approach to solving single-agent and adversarial graph search problems.

Hansson, Othar↗
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 NTRS2003

Scoping Planning Agents With Shared Models

In this paper we provide a formal framework to define the scope of planning agents based on a single declarative model. Having multiple agents sharing a single model provides numerous advantages that lead to reduced development costs and increase reliability of the system. We formally define planning in terms of extensions of an initial partial plan, and a set of flaws that make the plan unacceptable. A Flaw Filter (FF) allows us to identify those flaws relevant to an agent. Flaw filters motivate the Plan Identification Function (PIF), which specifies when an agent is is ready hand control to another agent for further work. PIFs define a set of plan extensions that can be generated from a model and a plan request. FFs and PIFs can be used to define the scope of agents without changing the model. We describe an implementation of PIFsand FFswithin the context of EUROPA, a constraint-based planning architecture, and show how it can be used to easily design many different agents.

Bedrax-Weiss, Tania↗
NASA NTRS1982

Chemical weathering and the Viking biology experiments on Mars

Since the publication of studies describing models that do not require external energy sources, new LR (labeled release) test standards module studies have been reported by Levin and Straat (1981). These studies impose important constraints on the various models, and they offer strong support for the 'frost-weathering' model. These findings are discussed here, and it is proposed that the activity of all three biology experiments can now adequately be explained as a consequence of a single active agent produced by contemporary chemical weathering. Some of the possible consequences and implications of contemporary weathering for the current state and evolution of the regolith and atmosphere are briefly discussed.

Huguenin, R. L.↗