Agent Architecture Research Articles

Retraction Note: An agent architecture for autonomous UAV flight control in object classification and recognition missions

Retraction Note: An agent architecture for autonomous UAV flight control in object classification and recognition missions

Chemical strategies for antisense antibiotics.

Antibacterial resistance is a severe threat to modern medicine and human health. To stay ahead of constantly-evolving bacteria we need to expand our arsenal of effective antibiotics. As such, antisense therapy is an attractive approach. The programmability allows to in principle target any RNA sequence within bacteria, enabling tremendous selectivity. In this Tutorial Review we provide guidelines for devising effective antibacterial antisense agents and offer a concise perspective for future research. We will review the chemical architectures of antibacterial antisense agents with a special focus on the delivery and target selection for successful antisense design. This Tutorial Review will strive to serve as an essential guide for antibacterial antisense technology development.

Autonomous Robot Task Execution in Flexible Manufacturing: Integrating PDDL and Behavior Trees in ARIAC 2023.

The Agile Robotics for Industrial Automation Competition (ARIAC) was established to advance flexible manufacturing, aiming to increase the agility of robotic assembly systems in unstructured and dynamic industrial environments. ARIAC 2023 introduced eight agility challenges involving faulty parts, flipped parts, faulty grippers, robot malfunctions, sensor blackouts, high-priority orders, insufficient parts, and human safety. Given the unpredictability of these scenarios, it is impractical to develop a specific strategy for each possible situation. To address these issues, this paper presents a hierarchical framework for autonomous robotic task generation and execution in dynamic scenarios. The framework is divided into a task level and an execution level. Initially, an immediate task management strategy is adopted at the task level, which reasonably decomposes dynamic tasks and allocates short-term tasks to the floor robot and ceiling robot. Later, at the execution level, each robot is designed with an agent architecture that combines PDDL planning with the quick response of behavior trees. Finally, the effectiveness and practicality of the proposed framework were thoroughly validated in ARIAC 2023.

Is Alignment Unsafe?

Yum (2024) argues that the widespread adoption of language agent architectures would likely increase the risk posed by AI by simplifying the process of aligning artificial systems with human values and thereby making it easier for malicious actors to use them to cause a variety of harms. Yum takes this to be an example of a broader phenomenon: progress on the alignment problem is likely to be net safety-negative because it makes artificial systems easier for malicious actors to control. I offer some reasons for skepticism about this surprising and pessimistic conclusion.

An Active Inference Agent for Modeling Human Translation Processes.

This paper develops an outline for a hierarchically embedded architecture of an artificial agent that models human translation processes based on principles of active inference (AIF) and predictive processing (PP). AIF and PP posit that the mind constructs a model of the environment which guides behavior by continually generating and integrating predictions and sensory input. The proposed model of the translation agent consists of three processing strata: a sensorimotor layer, a cognitive layer, and a phenomenal layer. Each layer consists of a network of states and transitions that interact on different time scales. Following the AIF framework, states are conditioned on observations which may originate from the environment and/or the embedded processing layer, while transitions between states are conditioned on actions that implement plans to optimize goal-oriented behavior. The AIF agent aims at simulating the variation in translational behavior under various conditions and to facilitate investigating the underlying mental mechanisms. It provides a novel framework for generating and testing new hypotheses of the translating mind.

Artificial Intelligent Agent Architecture and Clinical Decision-Making in the Healthcare Sector.

This paper examines the decision-making processes of physicians and intelligent agents within the healthcare sector, particularly focusing on their characteristics, architectures, and approaches. We provide a theoretical insight into the evolving role of artificial intelligence (AI) in healthcare, emphasizing its potential to address various healthcare challenges. Defining features of intelligent agents are explored, including their perceptual abilities and behavioral properties, alongside their architectural frameworks, ranging from reflex-based to general learning agents, and contrasted with the rational decision-making structure employed by physicians. Through data collection, hypothesis generation, testing, and reflection, physicians exhibit a nuanced approach informed by adaptability and contextual understanding. A comparative analysis between intelligent agents and physicians reveals both similarities and disparities, particularly in adaptability and contextual comprehension. While intelligent agents offer promise in enhancing clinical decisions, challenges with types of dataset biases pose significant hurdles. Informing and educating physicians about AI concepts can build trust and transparency in intelligent programs. Such efforts aim to leverage the strengths of both human and AI toward improving healthcare delivery and outcomes.

A method of using modern endpoint detection and response (EDR) systems to protect against complex attacks

The subject of the research in this article is the architecture of Endpoint Detection and Response and the EDR agent as their base parts in terms of mechanisms for detecting and countering complex attacks on information and communication systems (ICS). The aim of the work is to develop of method for improving the efficiency of using Endpoint Detection and Response (EDR) to reduce the risks of compromising ICS information, industrial, and infrastructure objects by effectively redistributing and utilizing the available EDR mechanisms, the cybersecurity team, and other resources available for implementing security measures in an enterprise, institution, or organization. The article addresses the following tasks: reviewing and analyzing existing EDR systems, analyzing the architecture of EDR solutions and EDR agents, the features of their use, the logic behind the construction of methods and mechanisms for detecting threats to the system from malicious actors and malicious code. The task of providing recommendations for the organization of ICS is also separately addressed in terms of the need to protect the entire ICS and its individual elements, as well as in terms of the available resources (the cybersecurity team, their qualifications and level of awareness of the architecture of EDR solutions) and means (available EDR system elements) for organizing protection. The following methods are used: modeling attack mechanisms, modeling attacker behavior. The following results were obtained: general and specific recommendations were formulated for optimizing the operation of EDR systems and ensuring the effective use of EDR system elements in the information and communication networks of enterprises, organizations, and institutions of various types and orientations depending on the available resources and the information requiring protection. Conclusions: The identified recommendations for the application of EDR mechanisms for protecting information systems and networks allow optimizing the costs of creating a protection infrastructure and implementing security measures, taking into account the characteristics of the available tools and the training and awareness of the cybersecurity team both in terms of response time to threats and the complexity and cost of performing protection tasks.

Circular (de)construction matchmaking: A matter of space and time

AbstractIndustrial symbiosis (IS) facilitates the transition toward a circular built environment. Following IS principles, multiple buildings can be symbiotically linked via closed‐loop material flows beyond the boundaries of individual projects. However, there are few IS matchmaking methods that support the identification of IS opportunities among multiple deconstruction and construction projects. This research develops an agent‐based model to fill this gap. The agent architecture is designed based on the concept of shearing layers. Circularity hubs are proposed to support IS matchmaking by allowing larger transportation ranges and keeping IS requests active for longer periods. The model's applicability is demonstrated through an industrial–urban symbiosis case in Enschede, the Netherlands. The model simulates the spatial–temporal dynamics of IS matchmaking as an emergent phenomenon under future scenarios. The results show operational evidence of IS matchmaking via the strategic implementation of circularity hubs. Overall, this research provides a new methodological perspective to explore the circularity in the built environment at scale.

A Review of Personalized Federated Reinforcement Learning

Reinforcement learning and federated learning both provide strong theoretical support for the study of artificial intelligence. In recent years, an emerging federated reinforcement learning paradigm has been proposed and widely studied and applied. However, in the federated reinforcement learning architecture, the environment, data type and device performance of different agents may be different, which is called heterogeneity. The existence of heterogeneity factors may lead to slow convergence speed of the algorithm, poor generalization, and suboptimal quality of the trained model. Therefore, how to solve the negative impact of the heterogeneity problem on model training has become a hot content of research, and the most important method is to train personalized models for clients. This paper introduces the theory of federated reinforcement learning, as well as methods to cope with heterogeneity in federated reinforcement learning, and provides an overview of the applications of federated reinforcement learning. We conclude the paper with a summary and future perspectives.

A domain-independent agent architecture for adaptive operation in evolving open worlds

1Angry Birds Competition. Model-based reasoning agents are ill-equipped to act in novel situations in which their model of the environment no longer sufficiently represents the world. We propose HYDRA, a framework for designing model-based agents operating in mixed discrete-continuous worlds that can autonomously detect when the environment has evolved from its canonical setup, understand how it has evolved, and adapt the agents' models to perform effectively. HYDRA is based upon PDDL+, a rich modeling language for planning in mixed, discrete-continuous environments. It augments the planning module with visual reasoning, task selection, and action execution modules for closed-loop interaction with complex environments. HYDRA implements a novel meta-reasoning process that enables the agent to monitor its own behavior from a variety of aspects. The process employs a diverse set of computational methods to maintain expectations about the agent's own behavior in an environment. Divergences from those expectations are useful in detecting when the environment has evolved and identifying opportunities to adapt the underlying models. HYDRA builds upon ideas from diagnosis and repair and uses a heuristics-guided search over model changes such that they become competent in novel conditions. The HYDRA framework has been used to implement novelty-aware agents for three diverse domains - CartPole++ (a higher dimension variant of a classic control problem), Science Birds (an IJCAI competition problem1), and PogoStick (a specific problem domain in Minecraft). We report empirical observations from these domains to demonstrate the efficacy of various components in the novelty meta-reasoning process.

Reflective Artificial Intelligence

As artificial intelligence (AI) technology advances, we increasingly delegate mental tasks to machines. However, today’s AI systems usually do these tasks with an unusual imbalance of insight and understanding: new, deeper insights are present, yet many important qualities that a human mind would have previously brought to the activity are utterly absent. Therefore, it is crucial to ask which features of minds have we replicated, which are missing, and if that matters. One core feature that humans bring to tasks, when dealing with the ambiguity, emergent knowledge, and social context presented by the world, is reflection. Yet this capability is completely missing from current mainstream AI. In this paper we ask what reflective AI might look like. Then, drawing on notions of reflection in complex systems, cognitive science, and agents, we sketch an architecture for reflective AI agents, and highlight ways forward.

An appraisal-based chain-of-emotion architecture for affective language model game agents.

The development of believable, natural, and interactive digital artificial agents is a field of growing interest. Theoretical uncertainties and technical barriers present considerable challenges to the field, particularly with regards to developing agents that effectively simulate human emotions. Large language models (LLMs) might address these issues by tapping common patterns in situational appraisal. In three empirical experiments, this study tests the capabilities of LLMs to solve emotional intelligence tasks and to simulate emotions. It presents and evaluates a new Chain-of-Emotion architecture for emotion simulation within video games, based on psychological appraisal research. Results show that it outperforms control LLM architectures on a range of user experience and content analysis metrics. This study therefore provides early evidence of how to construct and test affective agents based on cognitive processes represented in language models.

Dynamic knowledge graph applications for augmented built environments through “The World Avatar”

The proliferation of digital building models in recent years has led to a corresponding rise in specialised, non-interoperable models. These models impede sustainable developments by forming data silos that hinder cross-application data exchange and knowledge discovery processes. Although Semantic Web solutions hold promise in addressing these silos, current approaches primarily focus on developing novel ontologies, yielding similar outcomes. But it is unclear how these methodologies could support broader knowledge discovery processes and application requirements. This paper addresses these research challenges by introducing a dynamic knowledge graph as implemented within The World Avatar for interoperable building models. We demonstrate its value through two distinct applications in urban energy management and laboratory automation. The dynamic knowledge graph revolves around a comprehensive structured knowledge model constructed from ontologies and agents. Ontologies semantically annotate data and represent domain knowledge and their relationships with standardised definitions. When augmented with an agent architecture, the resulting knowledge model can align stakeholder perspectives and accommodate the dynamic and scalable nature of urban data. Moreover, the dynamic knowledge graph fosters innovative human-machine interactions through visualisation interfaces to augment knowledge discovery processes in the built environment for greater efficiencies and innovation. As the knowledge model expands, users gain access to a broader spectrum of private and public data sources and technologies, while reducing integration barriers. This is especially pertinent for smaller and less influential entities like municipal and local governments with limited resources, who can realise substantial benefits at reduced costs.

Reinforcing Cyber Defense: Generative AI Powered Intelligent Agent Architecture for Enhanced Security Operations

Reinforcing Cyber Defense: Generative AI Powered Intelligent Agent Architecture for Enhanced Security Operations

An agent architecture for expressive spatial knowledge and reasoning in land use modeling and simulations

This paper presents a new cognitive agent design approach integrating spatial knowledge representation and reasoning in agent-based modeling dedicated to land use simulations. A deep motivation for our agent-centric contribution is the ever-increasing development of spatially explicit agent simulation platforms. We build on this technological evolution and topology theory to endow the agent with human’s spatial representation and reasoning following a Belief–Desire–Intention architecture. A pilot implementation of the methodology with simulation experiments on a hunting model was carried out in GAMA platform to assess agent performances. Simulations display a consistent trend of animal population dynamics and also confirm a high model sensitivity to the integration of spatial knowledge and reasoning in agent-based models of human actor. These results demonstrate a successful implementation and the importance of spatial dimension in the expressive power and the validity of agent-based models. Future research efforts should therefore emphasize on designing human knowledge representation and incorporating learning abilities to improve models efficiency.

Empowering digital twins with large language models for global temporal feature learning

Digital Twin (DT), as an efficient technology for virtual-physical interaction, has demonstrated significant application potential in various industries. Intelligent agent-driven digital twin systems excel in analysis, decision-making, and control, making them highly suitable for manufacturing resource scheduling, diagnostic decision-making, and other requirements. However, current intelligent agents have notable deficiencies in adaptability, data utilization, and interpretability. This limitation undermines decision security and acceptability, creating barriers for user intervention. Therefore, this paper introduces a DT multi agent architecture driven by Large Language Models (LLMs). Agents perceive the characteristics of physical systems, particularly their temporal characteristics, by integrating data from various modalities. Multiple agents achieve insights through specific interaction mechanisms, while maintaining traceability. To showcase the advantages and characteristics of this architecture, we developed a typical application scenario for equipment maintenance. The effectiveness of each framework component was validated through ablation experiments. The experimental results suggest that the proposed framework holds promising and extensive application prospects.

End-to-end deep reinforcement learning for control of an autonomous underwater robot with an undulating propulsor

This paper focuses on the development and implementation of control algorithms for positioning an Autonomous Underwater Vehicle (AUV) with an undulating propulsor, using reinforcement learning methods. It provides an analysis and overview of works incorporating reinforcement learning methods such as Actor-only, Critic-only, and Actor-Critic. The paper primarily focuses on the Deep Deterministic Policy Gradient method and its implementation using deep neural networks to train the Actor-Critic agent. In the agent's architecture, a replay buffer and target neural networks were utilized to address the data correlation issue that induces training instability. An adaptive ar-chitecture was proposed for training the agent to force the robot to move from the initial point to any target point. Additionally, a random target point generator was incorporated at the training stage so as not to retrain the agent when the target points change. The training objective is to optimize the actor's policy by optimizing the critic and maximizing the reward function. Reward function is determined as the distance from the robot's center of mass to the target points. Consequently, the reward received by the agent increases when the robot gets closer to the target point and becomes maximal when the target point is reached with an acceptable error.

Implementing Cognitive Semantics of Autoepistemic Membership Statements: The Case of Categories with Prototypes

This article presents a model of an architecture of an artificial cognitive agent that performs the function of generating autoepistemic membership statements used to communicate beliefs about the belonging of an observed external object to a category with a prototype. The meaning of statements is described within the model by means of cognitive semantics. The presented proposal builds upon a pre-existing architecture and a semantic model designed for a simpler case of categories without a prototype. The main conclusion is that it is possible to develop an interactive cognitive agent capable of learning about categories with prototypes and producing autoepistemic membership statements fulfilling requirements of Rosch’s standard version of prototype semantics and satisfying pragmatic and logical rules for generating equivalents of these statements in natural languages. Detailed results include the following: an original proposal for an agent’s architecture, a model of an agent’s strategy of learning categories with a prototype, a scheme for determining the computational complexity of particular implementations of the learning strategy, definitions of cognitive semantics for particular cases of autoepistemic membership statements, and an analytical verification of properties of the proposed cognitive semantics. Finally, this article discusses the directions of further development and potential variants of the proposed architecture.

Amphiphilic pH-responsive ABC triblock copolymers via RAFT-mediated aqueous PISA and comparison with their A-b-(B-co-C) diblock copolymer counterparts

The influence of the macroRAFT agent architecture on the morphology of the self-assemblies obtained by aqueous RAFT dispersion polymerization in PISA is studied by comparing amphiphilic AC, A(B-co-C) diblocks and ABC triblock copolymers, constituted of N,N-dimethylacrylamide (DMAm = A), acrylic acid (AA = B) and N-cyanomethylacrylamide (CMAm = C) monomer units. The aim of this study is to better understand the impact of the presence of pH-sensitive acrylic acid units and their spatial distribution in block copolymers. Generally, the presence of a fully protonated intermediate block of PAA between the two blocks (based on PDMAm and PCMAm) favors the formation of higher-order morphologies. These ABC triblock copolymers assemblies are pH-responsive and can undergo morphological transitions through the controlled deprotonation or protonation of the PAA segment, but do not completely dissociate at high pH (i.e. when AA units are fully ionized) unlike their A(B-co-C) diblock counterparts, where the pH-responsive monomer unit is homogeneously distributed within the solvophobic block.

Growing an Embodied Generative Cognitive Agent

An evolutionary perspective on embodiment puts maintenance of physiology within a functional envelope as the brain’s base goal, with all other goals as refinements. Thus, all goals have physiological perturbation for their motivation and allostatic recovery as their signal of fulfillment. From this account, two entailments emerge. First, an object’s properties are not intrinsic to the object but a situated function of the morphology of the object and the affordances required by the goal. Second, categories do not exist without reference to some goal; they are constructed at the time of perception by blending prior conceptual knowledge to create an understanding of the perception with respect to the goal. Our thesis is that generative large language model (LLM) architectures are part of the solution to creating artificial organic-like cognitive architectures, but that LLMs as currently trained are generative only at a surface-level of behavior rather than deeper levels of cognition and, furthermore, that generative architectures must be coupled with an embodied cognitive agent architecture, which suggests both the additional levels at which generativity must operate and capabilities that the combined architecture must support.