Autonomous Agents Research Articles

Smart Cities towards Artificial Intelligence

The creation of smart cities has come a long way in recent years, especially in domains that use AI, such IoT-based components, security systems, smart living options, and environmental monitoring. These advancements seek to build surroundings that meet human requirements by utilizing cutting edge technologies such as automated parking, smart healthcare, and seismic monitoring systems. Even with these encouraging advancements, the current generation of smart cities is still unable to attain genuine autonomy, particularly when it comes to component-level autonomous situation management and decision-making. The transformational perspective proposed in this work views the constituent components of smart environments as self-governing agents with the ability to make decisions. The purpose of these autonomous agents is to improve security and smart living by being able to self-manage and react automatically to different situations. Such autonomous agents can be integrated to accomplish the goal of artificially intelligent smart cities. In smart cities, autonomous agents may behave pro-actively without human involvement by continually monitoring and analyzing data, anticipating possible problems, and taking proactive actions. In the context of smart healthcare, these agents may, for example, track a patient's vitals in real time, anticipate possible health emergencies, notify medical personnel, or even start emergency procedures. Autonomous agents may dynamically assign parking spots in the context of smart car parking based on real-time data, maximizing space use and minimizing traffic congestion. Agents that can autonomously manage emergency actions, such as directing people to safety and managing infrastructure to minimize damage, might be beneficial for earthquake detecting systems. As autonomous entities endowed with decision-making skills are included, we are moving closer to truly intelligent cities. Because of these agents' capacity for independent operation and quick response to changing circumstances, smart cities are able to offer a better quality of security and service while accommodating the changing demands of its residents. As a result, smart cities that have these intelligent agents installed will be more responsive, efficient, and resilient—ultimately capturing the spirit of artificial intelligence in urban planning.

Design Patterns for Building More Efficient Generative Autonomous Agents: A Survey

This survey paper explores the design patterns essential for constructing generative autonomous agents, a significant advancement over traditional LLM-based applications. These agents can independently plan, execute, and refine tasks, enabling them to overcome limitations such as the need for continuous human input and rules-based reasoning. I review four design patterns: reflection, planning, tool use, and multi-agent collaboration for building more efficient agents. Each is illustrated through a detailed, holistic analysis of a specific implementation technique.

Modeling Autonomous Vehicle Responses to Novel Observations Using Hierarchical Cognitive Representations Inspired Active Inference

Equipping autonomous agents for dynamic interaction and navigation is a significant challenge in intelligent transportation systems. This study aims to address this by implementing a brain-inspired model for decision making in autonomous vehicles. We employ active inference, a Bayesian approach that models decision-making processes similar to the human brain, focusing on the agent’s preferences and the principle of free energy. This approach is combined with imitation learning to enhance the vehicle’s ability to adapt to new observations and make human-like decisions. The research involved developing a multi-modal self-awareness architecture for autonomous driving systems and testing this model in driving scenarios, including abnormal observations. The results demonstrated the model’s effectiveness in enabling the vehicle to make safe decisions, particularly in unobserved or dynamic environments. The study concludes that the integration of active inference with imitation learning significantly improves the performance of autonomous vehicles, offering a promising direction for future developments in intelligent transportation systems.

Foundations & Trends in Multimodal Machine Learning: Principles, Challenges, and Open Questions

Multimodal machine learning is a vibrant multi-disciplinary research field that aims to design computer agents with intelligent capabilities such as understanding, reasoning, and learning through integrating multiple communicative modalities, including linguistic, acoustic, visual, tactile, and physiological messages. With the recent interest in video understanding, embodied autonomous agents, text-to-image generation, and multisensor fusion in application domains such as healthcare and robotics, multimodal machine learning has brought unique computational and theoretical challenges to the machine learning community given the heterogeneity of data sources and the interconnections often found between modalities. However, the breadth of progress in multimodal research has made it difficult to identify the common themes and open questions in the field. By synthesizing a broad range of application domains and theoretical frameworks from both historical and recent perspectives, this article is designed to provide an overview of the computational and theoretical foundations of multimodal machine learning. We start by defining three key principles of modality heterogeneity , connections , and interactions that have driven subsequent innovations, and propose a taxonomy of six core technical challenges: representation , alignment , reasoning , generation , transference , and quantification covering historical and recent trends. Recent technical achievements will be presented through the lens of this taxonomy, allowing researchers to understand the similarities and differences across new approaches. We end by motivating several open problems for future research as identified by our taxonomy.

Relative distances approach for multi-traveling salesmen problem

In this paper, we present a novel approach to addressing the Multi-Traveling Salesman Problem (MTSP), a complex optimization challenge involving the simultaneous fulfillment of multiple tasks such as cargo delivery and warehouse placement by multiple autonomous agents. In general there are two primary objectives to be achieved: minimizing the total path cost and minimizing the maximum cost incurred by agents, also known as makespan. While our primary focus centers on cost minimization, prioritizing this aspect alone often amplifies makespan. Our approach strikes a balance by ensuring that makespan remains within a reasonable threshold. Given the problem’s combinatorial nature, attaining cost-optimal solutions proves infeasible under current constraints. Hence, our emphasis shifts towards swift solution discovery to align with practical applicability. Consequently, we posit that the third pivotal objective is to curtail complexity and expedite solution acquisition, thereby enhancing the problem’s real-world viability. The Multi-Traveling Salesman Problem (MTSP) is typically approached in two distinct phases. In the initial phase, tasks are allocated to salesmen using various methods (such as K-Means or DBSCAN). The second phase involves determining the optimal routes for each salesman, a problem identical to the classic Traveling Salesman Problem (TSP). Our novel heuristic approach integrates these phases into a single method through our relative distance model. This model facilitates the easy addition and removal of tasks within the problem space and enables real-time scheduling.

Consensus Control of Multi-Agent System with Virtual Agents Considering Obstacle Avoidance

A multi-agent system (MAS) is a system whose overall behavior is determined by local interactions among multiple autonomous agents. Recently, research has been conducted on the application of MASs in real-world environments, in which the agents are assumed to be robots that drive on the ground, i.e., autonomous mobile robots, and acquire external environmental information using cameras. In such cases, the information that can be obtained by the agent is limited to the field of view (FOV) of the respective camera, and the overall graph structure is dynamic and time varying. In addition, because the FOV may be obstructed by obstacles during camera measurements, obstacle avoidance must be considered. In this study, we examined the MAS consensus problem considering the effects of a limited FOV obstructed by obstacles. Specifically, we propose a control method using virtual agents that considers obstacle avoidance based on funnel control. In addition, simulation study was performed to demonstrate the effectiveness of the proposed method for solving the MAS consensus problem in an environment with obstacles.

Simple autonomous agents can enhance creative semantic discovery by human groups

Innovation is challenging, and theory and experiments indicate that groups may be better able to identify and preserve innovations than individuals. But innovation within groups faces its own challenges, including groupthink and truncated diffusion. We performed experiments involving a game in which people search for ideas in various conditions: alone, in networked social groups, or in networked groups featuring autonomous agents (bots). The objective was to search a semantic space of 20,000 nouns with defined similarities for an arbitrary noun with the highest point value. Participants (N = 1875) were embedded in networks (n = 125) of 15 nodes to which we sometimes added 2 bots. The bots had 3 possible strategies: they shared a random noun generated by their immediate neighbors, or a noun most similar from among those identified, or a noun least similar. We first confirm that groups are better able to explore a semantic space than isolated individuals. Then we show that when bots that share the most similar noun operate in groups facing a semantic space that is relatively easy to navigate, group performance is superior. Simple autonomous agents with interpretable behavior can affect the capacity for creative discovery of human groups.

GeneWorker: An end-to-end robotic reinforcement learning approach with collaborative generator and worker networks

Reinforcement learning aided by the skill conception exhibits potent capabilities in guiding autonomous agents toward acquiring meaningful behaviors. However, in the current landscape of reinforcement learning, a skill is often merely a rudimentary abstraction of a sequence of primitive actions, serving as a component of the input to policy networks with fixed network parameters. This rigid methodology presents obstacles when attempting to integrate with burgeoning techniques such as meta-learning and large language models. To address this issue, we introduce a unique neural skill representation that abstracts the activation of neurons in each neural layer. Based on this, a novel end-to-end robotic reinforcement learning algorithm is proposed, in which two sub-networks, i.e., generator and worker networks, implement collaborative inferences via neural skills. Specifically, the generator produces a series of multi-spatial neural skills, providing efficient guidance for subsequent decision-making; by integrating these skills, the worker can determine its own network weights and biases to cope with various environmental conditions. Therefore, actions can be sampled with flexibly changeable network parameters through the collaboration between generator and worker networks. The experiments demonstrate that GeneWorker can achieve a mean success rate of over 90.67% on continuous robotic tasks and outperforms previous state-of-the-art methods by a minimum of 54% on the pick-and-place task.

Synthesis and Verification of Mission Plans for Multiple Autonomous Agents under Complex Road Conditions

Mission planning for multi-agent autonomous systems aims to generate feasible and optimal mission plans that satisfy given requirements. In this article, we propose a tool-supported mission-planning methodology that combines (i) a path-planning algorithm for synthesizing path plans that are safe in environments with complex road conditions, and (ii) a task-scheduling method for synthesizing task plans that schedule the tasks in the right and fastest order, taking into account the planned paths. The task-scheduling method is based on model checking, which provides means of automatically generating task execution orders that satisfy the requirements and ensure the correctness and efficiency of the plans by construction. We implement our approach in a tool named MALTA, which offers a user-friendly GUI for configuring mission requirements, a module for path planning, an integration with the model checker UPPAAL, and functions for automatic generation of formal models, and parsing of the execution traces of models. Experiments with the tool demonstrate its applicability and performance in various configurations of an industrial case study of an autonomous quarry. We also show the adaptability of our tool by employing it in a special case of an industrial case study.

AVA: Towards Autonomous Visualization Agents through Visual Perception‐Driven Decision‐Making

AbstractWith recent advances in multi‐modal foundation models, the previously text‐only large language models (LLM) have evolved to incorporate visual input, opening up unprecedented opportunities for various applications in visualization. Compared to existing work on LLM‐based visualization works that generate and control visualization with textual input and output only, the proposed approach explores the utilization of the visual processing ability of multi‐modal LLMs to develop Autonomous Visualization Agents (AVAs) that can evaluate the generated visualization and iterate on the result to accomplish user‐defined objectives defined through natural language. We propose the first framework for the design of AVAs and present several usage scenarios intended to demonstrate the general applicability of the proposed paradigm. Our preliminary exploration and proof‐of‐concept agents suggest that this approach can be widely applicable whenever the choices of appropriate visualization parameters require the interpretation of previous visual output. Our study indicates that AVAs represent a general paradigm for designing intelligent visualization systems that can achieve high‐level visualization goals, which pave the way for developing expert‐level visualization agents in the future.

An intelligent open trading system for on-demand delivery facilitated by deep Q network based reinforcement learning

On-demand delivery in urban areas has been growing rapidly in recent years. Nevertheless, on-demand delivery networks lack an efficient, sustainable, and environmentally friendly operative strategy. An open trading system equipped with on-line auctions provides an opportunity for increasing the efficiency of on-demand delivery systems. Reinforcement learning techniques that automate decision-making can facilitate the implementation of such complex and dynamic systems. This paper presents an on-line auction-based request trading platform embedded within an open trading system as a new scheme for carriers and shippers to trade on-demand delivery requests. The system is developed based on a multi-agent model, composed of carriers, shippers, and the on-line platform as autonomous agents. Deep Q network enabled reinforcement learning is used in the decision-making processes for the agents to optimise their behaviour in a dynamic environment. Numerical experiments conducted on the Melbourne metropolitan network demonstrate the effectiveness of the open trading system, which can provide benefits for all stakeholders involved in the on-demand delivery market as well as the entire system. The reinforcement learning enabled platform can gain more profit when there are more learning carriers. The results indicate that the intelligent open trading system with on-line auctions is a promising city logistics solution.

Automated ethical decision, value-ladenness, and the moral prior problem

AbstractPart of the literature on machine ethics and ethical artificial intelligence focuses on the idea of defining autonomous ethical agents able to make ethical choices and solve dilemmas. While ethical dilemmas often arise in situations characterized by uncertainty, the standard approach in artificial intelligence is to use rational choice theory and maximization of expected utility to model how algorithm should choose given uncertain outcomes. Motivated by the moral proxy problem, which proposes that the appraisal of ethical decisions varies depending on whether algorithms are considered to act as proxies for higher- or for lower-level agents, this paper introduces the moral prior problem, a limitation that, we believe, has been genuinely overlooked in the literature. In a nutshell, the moral prior problem amounts to the idea that, beyond the thesis of the value-ladenness of technologies and algorithms, automated ethical decisions are predetermined by moral priors during both conception and usage. As a result, automated decision procedures are insufficient to produce ethical choices or solve dilemmas, implying that we need to carefully evaluate what autonomous ethical agents are and can do, and what they aren’t and can’t.

A Taxonomy of Explanation Types and Need Indicators in Human–Agent Collaborations

In recent years, explanations have become a pressing matter in AI research. This development was caused by the increased use of black-box models and a realization of the importance of trustworthy AI. In particular, explanations are necessary for human–agent interactions to ensure that the user can trust the agent and that collaborations are effective. Human–agent interactions are complex social scenarios involving a user, an autonomous agent, and an environment or task with its own distinct properties. Thus, such interactions require a wide variety of explanations, which are not covered by the methods of a single AI discipline, such as computer vision or natural language processing. In this paper, we map out what types of explanations are important for human–agent interactions, surveying the field via a scoping review. In addition to the typical introspective explanation tackled by explainability researchers, we look at assistive explanations, aiming to support the user with their task. Secondly, we survey what causes the need for an explanation in the first place. We identify a variety of human–agent interaction-specific causes and categorize them by whether they are centered on the agent’s behavior, the user’s mental state, or an external entity. Our overview aims to guide robotics practitioners in designing agents with more comprehensive explanation-related capacities, considering different explanation types and the concrete times when explanations should be given.

Half mirror algorithm: a metaheuristic that hybridizes swarm intelligence and evolution-based system

This paper promotes a new metaheuristic called the half mirror algorithm (HMA). As its name suggests, HMA offers a new kind of mirroring search. HMA is developed by hybridizing swarm intelligence and the evolution system. Swarm intelligence is adopted by constructing several autonomous agents called swarms. On the other hand, the evolution system is adopted using arithmetic crossover based on a particular reference called a mirror. Four mirrors are used in HMA: the best swarm member, a randomly selected swarm member, the central point of the space, and the corresponding swarm member. During the confrontative assessment, HMA is confronted with average and subtraction-based optimization (ASBO), total interaction algorithm (TIA), walrus optimization algorithm (WaOA), coati optimization algorithm (COA), and clouded leopard optimization (CLO). The result shows that HMA is superior to ASBO, TIA, WaOA, COA, and CLO in 20, 19, 19, 20, and 20 out of 23 functions, respectively. Moreover, HMA has found the global optimal of eight functions. It means the superiority of HMA occurs in almost entire functions. In the future, the mirroring search can be combined with the guided and neighborhood search to construct a more powerful metaheuristic.

Can there be responsible AI without AI liability? Incentivizing generative AI safety through ex-post tort liability under the EU AI liability directive

Abstract In Europe, the governance discourse surrounding artificial intelligence (AI) has been predominantly centred on the AI Act, with a proliferation of books, certification courses, and discussions emerging even before its adoption. This narrow focus has overshadowed other crucial regulatory interventions that promise to fundamentally shape AI. This article highlights the proposed EU AI liability directive (AILD), the first attempt to harmonize general tort law in response to AI-related threats, addressing critical issues such as evidence discovery and causal links. As AI risks proliferate, this article argues for the necessity of a responsive system to adequately address AI harms as they arise. AI safety and responsible AI, central themes in current regulatory discussions, must be prioritized, with ex-post liability in tort playing a crucial role in achieving these objectives. This is particularly pertinent as AI systems become more autonomous and unpredictable, rendering the ex-ante risk assessments mandated by the AI Act insufficient. The AILD’s focus on fault and its limited scope is also inadequate. The proposed easing of the burden of proof for victims of AI, through enhanced discovery rules and presumptions of causal links, is insufficient in a context where Large Language Models exhibit unpredictable behaviours and humans increasingly rely on autonomous agents for complex tasks. Moreover, the AILD’s reliance on the concept of risk, inherited from the AI Act, is misplaced, as tort liability intervenes after the risk has materialized. However, the inherent risks in AI systems could justify EU harmonization of AI torts in the direction of strict liability. Bridging the liability gap will enhance AI safety and responsibility, better protect individuals from AI harms, and ensure that tort law remains a vital regulatory tool.

Энактивистский подход как выход из «ловушки когнитивизма»: реабилитация субъектности

The article substantiates the importance of the subjectness position in the study of cognitive processes. The subject cannot be either eliminated or considered as a mere “carrier” and “mediator” of cognitive abilities; it is necessary to study the whole variety of subjective ex­perience and the possibilities of its empirical fixation and measurability discovered by phe­nomenology. The article argues that the radical separation of cognitive processes from con­sciousness creates a kind of explanatory gap in scientific theories. According to the author, the “trap of cognitivism” arises if the cognitive process is viewed as the solution of predeter­mined tasks (assigned to the system from the outside), where the mind is essentially a cogni­tive unconscious, which deepens the gap between “computational” and phenomenological mind and leaves no room for subjectness. The article proves the productivity of the enac­tivist approach from the position of coherence of the concept of mind as a dynamic system and phenomenological explanation of the experience of subjectness, since autonomous agents create their own cognitive domains, which allows them to endow information with new meanings and perform situational actions relevant to the environment.

SayNav: Grounding Large Language Models for Dynamic Planning to Navigation in New Environments

Semantic reasoning and dynamic planning capabilities are crucial for an autonomous agent to perform complex navigation tasks in unknown environments. It requires a large amount of common-sense knowledge, that humans possess, to succeed in these tasks. We present SayNav, a new approach that leverages human knowledge from Large Language Models (LLMs) for efficient generalization to complex navigation tasks in unknown large-scale environments. SayNav uses a novel grounding mechanism, that incrementally builds a 3D scene graph of the explored environment as inputs to LLMs, for generating feasible and contextually appropriate high-level plans for navigation. The LLM-generated plan is then executed by a pre-trained low-level planner, that treats each planned step as a short-distance point-goal navigation sub-task. SayNav dynamically generates step-by-step instructions during navigation and continuously refines future steps based on newly perceived information. We evaluate SayNav on multi-object navigation (MultiON) task, that requires the agent to utilize a massive amount of human knowledge to efficiently search multiple different objects in an unknown environment. We also introduce a benchmark dataset for MultiON task employing ProcTHOR framework that provides large photo-realistic indoor environments with variety of objects. SayNav achieves state-of-the-art results and even outperforms an oracle based baseline with strong ground-truth assumptions by more than 8% in terms of success rate, highlighting its ability to generate dynamic plans for successfully locating objects in large-scale new environments. The code, benchmark dataset and demonstration videos are accessible at https://www.sri.com/ics/computer-vision/saynav.

Chitosan‐Decorated Alumina Hybrid Nanoparticles as Smart Scavengers of HF and Dissolved Transition Metals in Lithium‐Ion Batteries

AbstractLithium‐ion cells encompassing LiPF6 as the lithium salt and high‐voltage (LiNi0.5Mn1.5O4, LNMO) or high‐capacity (LiNi0.8Mn0.1Co0.1O2, NMC811) cathode materials are prone to transition metal (TM) dissolution caused by HF, whose formation is catalyzed by H2O traces. TM ions can shuttle to the anodic compartment, increasing the cell degradation rate. Accordingly, specific self‐healing strategies are helpful to develop scavengers able to eliminate HF and absorb TM ions so avoiding their shuttling. In this work, the fabrication and test of a bi‐functional, autonomous scavenging agent made of Al2O3 particles decorated is reported with chitosan. The nanometric Al2O3 core is a trap for HF by chemical bonding. The chitosan coating is acid‐sensitive, and the opening of such a capping layer is triggered in the presence of even small amounts of HF, leading to an efficient TM ion‐trapping. In addition, chitosan is biocompatible, biodegradable, and abundant, which is relevant for design‐for‐recycling scopes. With ≈200 ppm of water in the electrolyte, 12 wt% of scavenger causes, after 200 cycles at 1C, an increase of capacity retention from 73% to 88% for LNMO, and, impressively, from 46% to 84% for NMC811. This autonomous self‐healing mechanism is promising for application in next‐generation smart cells without requiring any external sensing.

RL-Based Sim2Real Enhancements for Autonomous Beach-Cleaning Agents

RL-Based Sim2Real Enhancements for Autonomous Beach-Cleaning Agents

Towards autonomous learning and optimisation in textile production: data-driven simulation approach for optimiser validation

AbstractThe textile industry is a traditional industry branch that remains highly relevant in Europe. The industry is under pressure to remain profitable in this high-wage region. As one promising approach, data-driven methods can be used for process optimisation in order to reduce waste, increase profitability and relieve mental burden on staff members. However, approaches from research rarely get adopted into practice. We identify the high dimensionality of textile production processes leading to high model uncertainty as well as an incomplete problem formulation as the two main problems. We argue that some form of an autonomous learning agent can address this challenge, when it safely explores advantageous, unknown new settings by interacting with the process. Our main goal is to facilitate the adoption of promising research into practical applications. The main contributions of this paper include the derivation and formulation of a probabilistic optimisation problem for high-dimensional, stationary production processes. We also create a highly adaptable simulation of the textile carded nonwovens production process in Python that implements the optimisation problem. Economic and technical behavior of the process is approximated using both Gaussian Process Regression (GPR) models trained with industrial data as well as physics-motivated explicit models. This ’simulation first’-approach makes the development of autonomous learning agents for practical applications feasible because it allows for cheap testing and validation before physical trials. Future work will include the comparison of the performance of different agent approaches.