Autonomy, a key property associated with the agent, is an important topic in the current research of the agent theory. Although no definition of the agent autonomy is universally accepted, an important aspect of the a...Autonomy, a key property associated with the agent, is an important topic in the current research of the agent theory. Although no definition of the agent autonomy is universally accepted, an important aspect of the agent autonomy is the decision-making capability of the agents. This paper investigates the autonomy of the agent, presents a framework for autonomous agent and discusses its decision-making process. Started with introducing a language for representing autonomous agent, a framework is proposed for modeling autonomous agent based on a BDI model and the situation calculus. Finally, a kind of decision-making process of the autonomous agent is presented.展开更多
This paper considers the formation control problem of multi-agent systems in a distributed fashion. Two cases of the information propagating topologies among multiple agents, characterized by graphics model, are consi...This paper considers the formation control problem of multi-agent systems in a distributed fashion. Two cases of the information propagating topologies among multiple agents, characterized by graphics model, are considered. One is fixed topology. The other is switching topology which represents the limited and less reliable information exchange. The local formation control strategies established in this paper are based on a simple modification of the existing consensus control strategies. Moreover, some existing convergence conditions are shown to be a special case of our model even in the continuous-time consensus case. Therefore, the results of this paper extend the existing results about the consensus problem.展开更多
Autonomous agents have long been a research focus in academic and industry communities.Previous research often focuses on training agents with limited knowledge within isolated environments,which diverges significantl...Autonomous agents have long been a research focus in academic and industry communities.Previous research often focuses on training agents with limited knowledge within isolated environments,which diverges significantly from human learning processes,and makes the agents hard to achieve human-like decisions.Recently,through the acquisition of vast amounts of Web knowledge,large language models(LLMs)have shown potential in human-level intelligence,leading to a surge in research on LLM-based autonomous agents.In this paper,we present a comprehensive survey of these studies,delivering a systematic review of LLM-based autonomous agents from a holistic perspective.We first discuss the construction of LLM-based autonomous agents,proposing a unified framework that encompasses much of previous work.Then,we present a overview of the diverse applications of LLM-based autonomous agents in social science,natural science,and engineering.Finally,we delve into the evaluation strategies commonly used for LLM-based autonomous agents.Based on the previous studies,we also present several challenges and future directions in this field.展开更多
This study provides a systematic analysis of the resource-consuming training of deep reinforcement-learning (DRL) agents for simulated low-speed automated driving (AD). In Unity, this study established two case studie...This study provides a systematic analysis of the resource-consuming training of deep reinforcement-learning (DRL) agents for simulated low-speed automated driving (AD). In Unity, this study established two case studies: garage parking and navigating an obstacle-dense area. Our analysis involves training a path-planning agent with real-time-only sensor information. This study addresses research questions insufficiently covered in the literature, exploring curriculum learning (CL), agent generalization (knowledge transfer), computation distribution (CPU vs. GPU), and mapless navigation. CL proved necessary for the garage scenario and beneficial for obstacle avoidance. It involved adjustments at different stages, including terminal conditions, environment complexity, and reward function hyperparameters, guided by their evolution in multiple training attempts. Fine-tuning the simulation tick and decision period parameters was crucial for effective training. The abstraction of high-level concepts (e.g., obstacle avoidance) necessitates training the agent in sufficiently complex environments in terms of the number of obstacles. While blogs and forums discuss training machine learning models in Unity, a lack of scientific articles on DRL agents for AD persists. However, since agent development requires considerable training time and difficult procedures, there is a growing need to support such research through scientific means. In addition to our findings, we contribute to the R&D community by providing our environment with open sources.展开更多
Enhancing Autonomous Decision-Making (ADM) for unmanned combat aerial vehicle formations in beyond-visual-range air combat is pivotal for future battlefields, whereas the predominant reinforcement learning technique f...Enhancing Autonomous Decision-Making (ADM) for unmanned combat aerial vehicle formations in beyond-visual-range air combat is pivotal for future battlefields, whereas the predominant reinforcement learning technique for ADM has been proven to be inadequately fitting complex tactical Unit Coordination (UC), limiting the integrity of decision-making for formations. This study proposes a knowledge-enhanced ADM method, with a focus on UC, to elevate formation combat effectiveness. The main innovation is integrating data mining technique with tactical knowledge mining and integration. Foremost, based on Frequent Event Arrangement Mining (FEAM) theory, a cross-channel UC knowledge mining method is designed by introducing data flow, which is capable of capturing dynamic coordinative action sequences. Then, a dual-mode knowledge integration method is proposed by employing the Graph Attention Network (GAT) and attenuated structural similarity, bolstering the interplay between autonomous UC tactics fitting and knowledge injection. The experimental results demonstrate that the algorithm surpasses the existing methods, providing more strategic maneuver trajectories and a win rate of more than 90% in different scenarios. The method is promising to augment the autonomous operational capabilities of unmanned formations and drive the evolution of combat effectiveness.展开更多
The IBM Agent Building and Learning Environment(ABLE) provides a lightweight Java^(TM) agent frame- work,a comprehensive JavaBeansTM library of intelligent software components,a set of development and test tools, and ...The IBM Agent Building and Learning Environment(ABLE) provides a lightweight Java^(TM) agent frame- work,a comprehensive JavaBeansTM library of intelligent software components,a set of development and test tools, and an agent platform.After the introduction to ABLE,classes and interfaces in the ABLE agent framework were put forward.At last an autonomic agent that is an ABLE-based architecture for incrementally building autonomic systems was discussed.展开更多
Little by little, we are entering the new era, intelligent interfaces are absorbing us more and more every day, and artificial intelligence makes its presence in a stealthy way. Virtual humans that represent an evolut...Little by little, we are entering the new era, intelligent interfaces are absorbing us more and more every day, and artificial intelligence makes its presence in a stealthy way. Virtual humans that represent an evolution of autonomous virtual agents;they are computer programs and in the future capable of carrying out different activities in certain environments. They will give the illusion of being human;they will have a body, and they will be immersed in an environment. They will have a set of senses that will allow them: 1) Sensations and therefore associated expressions;2) Communication;3) Learning;4) Remembering events, among others. By integrating the above, they will have a personality and autonomy, so they will be able to plan with respect to objectives;allowing them to decide and take actions with their body, in other words, they will count on awareness. The applications will be focused on environments that they will inhabit, or as interfaces that will interact with other systems. The application domains will be multiple;one of them being education. This article shows the design of OANNA like an avatar with the role of pedagogical agent. It was modeled as an affective-cognitive structure related to the teaching-learning process linked to a pedagogical agent that represents the interface of an artilect. OANNA, has the necessary animations for intervention within the teaching-learning process.展开更多
A novel distributed model predictive control scheme based on dynamic integrated system optimization and parameter estimation （DISOPE） was proposed for nonlinear cascade systems under network environment. Under the d...A novel distributed model predictive control scheme based on dynamic integrated system optimization and parameter estimation （DISOPE） was proposed for nonlinear cascade systems under network environment. Under the distributed control structure, online optimization of the cascade system was composed of several cascaded agents that can cooperate and exchange information via network communication. By iterating on modified distributed linear optimal control problems on the basis of estimating parameters at every iteration the correct optimal control action of the nonlinear model predictive control problem of the cascade system could be obtained, assuming that the algorithm was convergent. This approach avoids solving the complex nonlinear optimization problem and significantly reduces the computational burden. The simulation results of the fossil fuel power unit are illustrated to verify the effectiveness and practicability of the proposed algorithm.展开更多
Modifications to an image feature extraction approach involving evolutionary computation and autonomous agents are proposed. The described algorithm allows extraction of features with certain specified characteristics...Modifications to an image feature extraction approach involving evolutionary computation and autonomous agents are proposed. The described algorithm allows extraction of features with certain specified characteristics, while omitting other undesirable details in the image. Experimental results are presented with remarks.展开更多
This paper demonstrates the potential role of autonomous agents in economic theory.We first dispatch autonomous agents,built by genetic programming,to double auction markets.We then study the bargaining strategies,dis...This paper demonstrates the potential role of autonomous agents in economic theory.We first dispatch autonomous agents,built by genetic programming,to double auction markets.We then study the bargaining strategies,discovered by them,and from there,an autonomous-agent-inspired economic theory with regard to the optimal procrastination is derived.展开更多
Large Language Models(LLMs),such as ChatGPT,demonstrate a strong understanding of human natural language and have been explored and applied in various fields,including reasoning,creative writing,code generation,transl...Large Language Models(LLMs),such as ChatGPT,demonstrate a strong understanding of human natural language and have been explored and applied in various fields,including reasoning,creative writing,code generation,translation,and information retrieval.By adopting LLM as the reasoning core,we introduce Autonomous GIS as an AI-powered geographic information system(GIS)that leverages the LLM’s general abilities in natural language understanding,reasoning,and coding for addressing spatial problems with automatic spatial data collection,analysis,and visualization.We envision that autonomous GIS will need to achieve five autonomous goals:self-generating,self-organizing,selfverifying,self-executing,and self-growing.We developed a prototype system called LLM-Geo using the GPT-4 API,demonstrating what an autonomous GIS looks like and how it delivers expected results without human intervention using three case studies.For all case studies,LLMGeo returned accurate results,including aggregated numbers,graphs,and maps..Although still in its infancy and lacking several important modules such as logging and code testing,LLM-Geo demonstrates a potential path toward the next-generation AI-powered GIS.We advocate for the GIScience community to devote more efforts to the research and development of autonomous GIS,making spatial analysis easier,faster,and more accessible to a broader audience.展开更多
Purpose–The purpose of this paper is to propose a layered adjustable autonomy(LAA)as a dynamically adjustable autonomy model for a multi-agent system.It is mainly used to efficiently manage humans’and agents’shared...Purpose–The purpose of this paper is to propose a layered adjustable autonomy(LAA)as a dynamically adjustable autonomy model for a multi-agent system.It is mainly used to efficiently manage humans’and agents’shared control of autonomous systems and maintain humans’global control over the agents.Design/methodology/approach–The authors apply the LAA model in an agent-based autonomous unmanned aerial vehicle(UAV)system.The UAV system implementation consists of two parts:software and hardware.The software part represents the controller and the cognitive,and the hardware represents the computing machinery and the actuator of the UAV system.The UAV system performs three experimental scenarios of dance,surveillance and search missions.The selected scenarios demonstrate different behaviors in order to create a suitable test plan and ensure significant results.Findings–The results of the UAV system tests prove that segregating the autonomy of a system as multidimensional and adjustable layers enables humans and/or agents to perform actions at convenient autonomy levels.Hence,reducing the adjustable autonomy drawbacks of constraining the autonomy of the agents,increasing humans’workload and exposing the system to disturbances.Originality/value–The application of the LAA model in a UAV manifests the significance of implementing dynamic adjustable autonomy.Assessing the autonomy within three phases of agents run cycle(taskselection,actions-selection and actions-execution)is an original idea that aims to direct agents’autonomy toward performance competency.The agents’abilities are well exploited when an incompetent agent switches with a more competent one.展开更多
文摘Autonomy, a key property associated with the agent, is an important topic in the current research of the agent theory. Although no definition of the agent autonomy is universally accepted, an important aspect of the agent autonomy is the decision-making capability of the agents. This paper investigates the autonomy of the agent, presents a framework for autonomous agent and discusses its decision-making process. Started with introducing a language for representing autonomous agent, a framework is proposed for modeling autonomous agent based on a BDI model and the situation calculus. Finally, a kind of decision-making process of the autonomous agent is presented.
基金the National Natural Science Foundation of China (No.60674071).
文摘This paper considers the formation control problem of multi-agent systems in a distributed fashion. Two cases of the information propagating topologies among multiple agents, characterized by graphics model, are considered. One is fixed topology. The other is switching topology which represents the limited and less reliable information exchange. The local formation control strategies established in this paper are based on a simple modification of the existing consensus control strategies. Moreover, some existing convergence conditions are shown to be a special case of our model even in the continuous-time consensus case. Therefore, the results of this paper extend the existing results about the consensus problem.
基金the National Natural Science Foundation of China(Grant No.62102420)the Beijing Outstanding Young Scientist Program(No.BJJWZYJH012019100020098)。
文摘Autonomous agents have long been a research focus in academic and industry communities.Previous research often focuses on training agents with limited knowledge within isolated environments,which diverges significantly from human learning processes,and makes the agents hard to achieve human-like decisions.Recently,through the acquisition of vast amounts of Web knowledge,large language models(LLMs)have shown potential in human-level intelligence,leading to a surge in research on LLM-based autonomous agents.In this paper,we present a comprehensive survey of these studies,delivering a systematic review of LLM-based autonomous agents from a holistic perspective.We first discuss the construction of LLM-based autonomous agents,proposing a unified framework that encompasses much of previous work.Then,we present a overview of the diverse applications of LLM-based autonomous agents in social science,natural science,and engineering.Finally,we delve into the evaluation strategies commonly used for LLM-based autonomous agents.Based on the previous studies,we also present several challenges and future directions in this field.
文摘This study provides a systematic analysis of the resource-consuming training of deep reinforcement-learning (DRL) agents for simulated low-speed automated driving (AD). In Unity, this study established two case studies: garage parking and navigating an obstacle-dense area. Our analysis involves training a path-planning agent with real-time-only sensor information. This study addresses research questions insufficiently covered in the literature, exploring curriculum learning (CL), agent generalization (knowledge transfer), computation distribution (CPU vs. GPU), and mapless navigation. CL proved necessary for the garage scenario and beneficial for obstacle avoidance. It involved adjustments at different stages, including terminal conditions, environment complexity, and reward function hyperparameters, guided by their evolution in multiple training attempts. Fine-tuning the simulation tick and decision period parameters was crucial for effective training. The abstraction of high-level concepts (e.g., obstacle avoidance) necessitates training the agent in sufficiently complex environments in terms of the number of obstacles. While blogs and forums discuss training machine learning models in Unity, a lack of scientific articles on DRL agents for AD persists. However, since agent development requires considerable training time and difficult procedures, there is a growing need to support such research through scientific means. In addition to our findings, we contribute to the R&D community by providing our environment with open sources.
文摘Enhancing Autonomous Decision-Making (ADM) for unmanned combat aerial vehicle formations in beyond-visual-range air combat is pivotal for future battlefields, whereas the predominant reinforcement learning technique for ADM has been proven to be inadequately fitting complex tactical Unit Coordination (UC), limiting the integrity of decision-making for formations. This study proposes a knowledge-enhanced ADM method, with a focus on UC, to elevate formation combat effectiveness. The main innovation is integrating data mining technique with tactical knowledge mining and integration. Foremost, based on Frequent Event Arrangement Mining (FEAM) theory, a cross-channel UC knowledge mining method is designed by introducing data flow, which is capable of capturing dynamic coordinative action sequences. Then, a dual-mode knowledge integration method is proposed by employing the Graph Attention Network (GAT) and attenuated structural similarity, bolstering the interplay between autonomous UC tactics fitting and knowledge injection. The experimental results demonstrate that the algorithm surpasses the existing methods, providing more strategic maneuver trajectories and a win rate of more than 90% in different scenarios. The method is promising to augment the autonomous operational capabilities of unmanned formations and drive the evolution of combat effectiveness.
文摘The IBM Agent Building and Learning Environment(ABLE) provides a lightweight Java^(TM) agent frame- work,a comprehensive JavaBeansTM library of intelligent software components,a set of development and test tools, and an agent platform.After the introduction to ABLE,classes and interfaces in the ABLE agent framework were put forward.At last an autonomic agent that is an ABLE-based architecture for incrementally building autonomic systems was discussed.
文摘Little by little, we are entering the new era, intelligent interfaces are absorbing us more and more every day, and artificial intelligence makes its presence in a stealthy way. Virtual humans that represent an evolution of autonomous virtual agents;they are computer programs and in the future capable of carrying out different activities in certain environments. They will give the illusion of being human;they will have a body, and they will be immersed in an environment. They will have a set of senses that will allow them: 1) Sensations and therefore associated expressions;2) Communication;3) Learning;4) Remembering events, among others. By integrating the above, they will have a personality and autonomy, so they will be able to plan with respect to objectives;allowing them to decide and take actions with their body, in other words, they will count on awareness. The applications will be focused on environments that they will inhabit, or as interfaces that will interact with other systems. The application domains will be multiple;one of them being education. This article shows the design of OANNA like an avatar with the role of pedagogical agent. It was modeled as an affective-cognitive structure related to the teaching-learning process linked to a pedagogical agent that represents the interface of an artilect. OANNA, has the necessary animations for intervention within the teaching-learning process.
基金This work was supportedbytheNationalNaturalScienceFoundationofChina(No.60474051),theProgramforNewCenturyExcellentTalentsinUniversityofChina(NCET),andtheSpecializedResearchFundfortheDoctoralProgramofHigherEducationofChina(No.20020248028).
文摘A novel distributed model predictive control scheme based on dynamic integrated system optimization and parameter estimation （DISOPE） was proposed for nonlinear cascade systems under network environment. Under the distributed control structure, online optimization of the cascade system was composed of several cascaded agents that can cooperate and exchange information via network communication. By iterating on modified distributed linear optimal control problems on the basis of estimating parameters at every iteration the correct optimal control action of the nonlinear model predictive control problem of the cascade system could be obtained, assuming that the algorithm was convergent. This approach avoids solving the complex nonlinear optimization problem and significantly reduces the computational burden. The simulation results of the fossil fuel power unit are illustrated to verify the effectiveness and practicability of the proposed algorithm.
文摘Modifications to an image feature extraction approach involving evolutionary computation and autonomous agents are proposed. The described algorithm allows extraction of features with certain specified characteristics, while omitting other undesirable details in the image. Experimental results are presented with remarks.
基金grateful to the participants and two anonymous referees for the comments received.The NSC grants NSC 98-2911-I-004-007,NSC 98-2410-H-004-045-MY3 are also gratefully acknowledged.
文摘This paper demonstrates the potential role of autonomous agents in economic theory.We first dispatch autonomous agents,built by genetic programming,to double auction markets.We then study the bargaining strategies,discovered by them,and from there,an autonomous-agent-inspired economic theory with regard to the optimal procrastination is derived.
文摘Large Language Models(LLMs),such as ChatGPT,demonstrate a strong understanding of human natural language and have been explored and applied in various fields,including reasoning,creative writing,code generation,translation,and information retrieval.By adopting LLM as the reasoning core,we introduce Autonomous GIS as an AI-powered geographic information system(GIS)that leverages the LLM’s general abilities in natural language understanding,reasoning,and coding for addressing spatial problems with automatic spatial data collection,analysis,and visualization.We envision that autonomous GIS will need to achieve five autonomous goals:self-generating,self-organizing,selfverifying,self-executing,and self-growing.We developed a prototype system called LLM-Geo using the GPT-4 API,demonstrating what an autonomous GIS looks like and how it delivers expected results without human intervention using three case studies.For all case studies,LLMGeo returned accurate results,including aggregated numbers,graphs,and maps..Although still in its infancy and lacking several important modules such as logging and code testing,LLM-Geo demonstrates a potential path toward the next-generation AI-powered GIS.We advocate for the GIScience community to devote more efforts to the research and development of autonomous GIS,making spatial analysis easier,faster,and more accessible to a broader audience.
文摘Purpose–The purpose of this paper is to propose a layered adjustable autonomy(LAA)as a dynamically adjustable autonomy model for a multi-agent system.It is mainly used to efficiently manage humans’and agents’shared control of autonomous systems and maintain humans’global control over the agents.Design/methodology/approach–The authors apply the LAA model in an agent-based autonomous unmanned aerial vehicle(UAV)system.The UAV system implementation consists of two parts:software and hardware.The software part represents the controller and the cognitive,and the hardware represents the computing machinery and the actuator of the UAV system.The UAV system performs three experimental scenarios of dance,surveillance and search missions.The selected scenarios demonstrate different behaviors in order to create a suitable test plan and ensure significant results.Findings–The results of the UAV system tests prove that segregating the autonomy of a system as multidimensional and adjustable layers enables humans and/or agents to perform actions at convenient autonomy levels.Hence,reducing the adjustable autonomy drawbacks of constraining the autonomy of the agents,increasing humans’workload and exposing the system to disturbances.Originality/value–The application of the LAA model in a UAV manifests the significance of implementing dynamic adjustable autonomy.Assessing the autonomy within three phases of agents run cycle(taskselection,actions-selection and actions-execution)is an original idea that aims to direct agents’autonomy toward performance competency.The agents’abilities are well exploited when an incompetent agent switches with a more competent one.
