Design and Development of Multi-Agent Reinforcement Learning Intelligence on the Robotarium Platform for Embedded System Applications DOI Open Access
Lorenzo Canese, G.C. Cardarilli, Mohammad Mahdi Dehghan Pir

et al.

Electronics, Journal Year: 2024, Volume and Issue: 13(10), P. 1819 - 1819

Published: May 8, 2024

This research explores the use of Q-Learning for real-time swarm (Q-RTS) multi-agent reinforcement learning (MARL) algorithm robotic applications. study investigates efficacy Q-RTS in reducing convergence time to a satisfactory movement policy through successful implementation four and eight trained agents. has been shown significantly reduce search terms training iterations, from almost million iterations with one agent 650,000 agents 500,000 The scalability was addressed by testing it on several agents’ configurations. A central focus placed design sophisticated reward function, considering various postures their critical role optimizing Q-learning algorithm. Additionally, this delved into robustness agents, revealing ability adapt dynamic environmental changes. findings have broad implications improving efficiency adaptability systems applications such as IoT embedded systems. tested implemented using Georgia Tech Robotarium platform, showing its feasibility above-mentioned

Language: Английский

Robotics in construction: A critical review of the reinforcement learning and imitation learning paradigms DOI Creative Commons
Juan Manuel Dávila Delgado, Lukumon O. Oyedele

Advanced Engineering Informatics, Journal Year: 2022, Volume and Issue: 54, P. 101787 - 101787

Published: Oct. 1, 2022

The reinforcement and imitation learning paradigms have the potential to revolutionise robotics. Many successful developments been reported in literature; however, these approaches not explored widely robotics for construction. objective of this paper is consolidate, structure, summarise research knowledge at intersection robotics, learning, A two-strand approach literature review was employed. bottom-up analyse detail a selected number relevant publications, top-down which large papers were analysed identify common themes trends. This study found that on construction has increased significantly since 1980s, terms publications. Also, lacks development dedicated systems, limits their effectiveness. Moreover, unlike manufacturing, construction's unstructured dynamic characteristics are major challenge approaches. provides very useful starting point understating by (i) identifying strengths limitations approaches, (ii) contextualising problem; both will aid kick-start subject or boost existing efforts.

Language: Английский

Citations

43
Internet of robotic things for mobile robots: Concepts, technologies, challenges, applications, and future directions DOI Creative Commons
Homayun Kabir, Mau‐Luen Tham, Yoong Choon Chang

et al.

Digital Communications and Networks, Journal Year: 2023, Volume and Issue: 9(6), P. 1265 - 1290

Published: May 29, 2023

Nowadays, Multi Robotic System (MRS) consisting of different robot shapes, sizes and capabilities has received significant attention from researchers are being deployed in a variety real-world applications. From sensors actuators improved by communication technologies to powerful computing systems utilizing advanced Artificial Intelligence (AI) algorithms have rapidly driven the development MRS, so Internet Things (IoT) MRS become new topic, namely Robots (IoRT). This paper summarises comprehensive survey state-of-the-art for mobile robots, including general architecture, benefits, challenges, practical applications, future research directions. In addition, remarkable i) multi-robot navigation, ii) network routing protocols communications, iii) coordination among robots as well data analysis via external (cloud, fog, edge, edge-cloud) merged with IoRT architecture according their applicability. Moreover, security is long-term challenge because various attack vectors, flaws, vulnerabilities. Security threats, attacks, existing solutions based on architectures also under scrutiny. identification environmental situations that crucial all types such detection objects, human, obstacles, critically reviewed. Finally, directions given analyzing challenges robots.

Language: Английский

Citations

40
A probabilistic deep reinforcement learning approach for optimal monitoring of a building adjacent to deep excavation DOI
Yue Pan,

Jianjun Qin,

Limao Zhang

et al.

Computer-Aided Civil and Infrastructure Engineering, Journal Year: 2023, Volume and Issue: 39(5), P. 656 - 678

Published: May 12, 2023

Abstract During a deep excavation project, monitoring the structural health of adjacent buildings is crucial to ensure safety. Therefore, this study proposes novel probabilistic reinforcement learning (PDRL) framework optimize plan minimize cost and excavation‐induced risk. First, Bayesian‐bi‐directional general regression neural network built as model describe relationship between ground settlement foundation pit safety state building, along with actions in dynamic manner. Subsequently, double Q‐network method, which can capture realistic features management problem, trained form closed decision loop for continuous strategies. Finally, proposed PDRL approach applied real‐world case No. 14 Shanghai Metro. This estimate time‐variant probability damage occurrence maintenance update building. According strategy via PDRL, begins middle stage rather than on first day project if there full confidence quality data. When uncertainty level data rises, starting might shift an earlier date. It worth noting that method adequately robust address uncertainties embedded environment model, thus contributing optimizing achieving cost‐effectiveness risk mitigation.

Language: Английский

Citations

30
Machine Learning for Tactile Perception: Advancements, Challenges, and Opportunities DOI Creative Commons

Zhixian Hu,

Lan Lin,

Waner Lin

et al.

Advanced Intelligent Systems, Journal Year: 2023, Volume and Issue: 5(7)

Published: March 15, 2023

The past decades have seen the rapid development of tactile sensors in material, fabrication, and mechanical structure design. advancement has heightened expectation sensor functions, thus put forward a higher demand for data processing. However, conventional analysis techniques not kept pace with still suffer from some severe drawbacks, like cumbersome models, poor efficiency, expensive costs. Machine learning, its prominent ability big fast processing speed, can offer many possibilities analysis. Herein, machine learning employed signals are reviewed. Supervised unsupervised analog covered, spike summarized. Furthermore, applications robotic perception human activity monitoring presented. Finally, current challenges future prospects sensors, data, algorithms, benchmarks discussed.

Language: Английский

Citations

25
Machine learning to predict the production of bio-oil, biogas, and biochar by pyrolysis of biomass: a review DOI
Kapil Khandelwal, Sonil Nanda, Ajay K. Dalai

et al.

Environmental Chemistry Letters, Journal Year: 2024, Volume and Issue: unknown

Published: Sept. 5, 2024

Language: Английский

Citations

11
Study of Q-learning and deep Q-network learning control for a rotary inverted pendulum system DOI Creative Commons
Zied Ben Hazem

Deleted Journal, Journal Year: 2024, Volume and Issue: 6(2)

Published: Feb. 2, 2024

Abstract The rotary inverted pendulum system (RIPS) is an underactuated mechanical with highly nonlinear dynamics and it difficult to control a RIPS using the classic models. In last few years, reinforcement learning (RL) has become popular method. RL powerful potential systems high non-linearity complex dynamics, such as RIPS. Nevertheless, for not been well studied there limited research on development evaluation of this paper, algorithms are developed swing-up stabilization single-link (SLRIP) compared methods PID LQR. A physical model SLRIP created MATLAB/Simscape Toolbox, used dynamic simulation in MATLAB/Simulink train agents. An agent trainer Q-learning (QL) deep Q-network (DQNL) proposed data training. Furthermore, actions actuating horizontal arm states angles velocities arm. reward computed according zero when attends upright position. without understanding classical controllers implement agent. Finally, outcome indicates effectiveness QL DQNL conventional LQR controllers.

Language: Английский

Citations

10
Examination of ChatGPT’s Performance as a Data Analysis Tool DOI
Duygu Koçak

Educational and Psychological Measurement, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 3, 2025

This study examines the performance of ChatGPT, developed by OpenAI and widely used as an AI-based conversational tool, a data analysis tool through exploratory factor (EFA). To this end, simulated were generated under various conditions, including normal distribution, response category, sample size, test length, loading, measurement models. The analyzed using ChatGPT-4o twice with 1-week interval same prompt, results compared those obtained R code. In analysis, Kaiser–Meyer–Olkin (KMO) value, total variance explained, number factors estimated empirical Kaiser criterion, Hull method, Kaiser–Guttman well loadings, calculated. findings from ChatGPT at two different times found to be consistent R. Overall, demonstrated good for steps that require only computational decisions without involving researcher judgment or theoretical evaluation (such KMO, loadings). However, multidimensional structures, although was across analyses, biases observed, suggesting researchers should exercise caution in such decisions.

Language: Английский

Citations

1
FTR‐Bench: Benchmarking Deep Reinforcement Learning for Flipper‐Track Robot Control DOI Open Access
Hongchuan Zhang, Junkai Ren, Junhao Xiao

et al.

Journal of Field Robotics, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 5, 2025

ABSTRACT Tracked robots equipped with flippers and sensors are extensively employed in outdoor search rescue scenarios. However, achieving precise motion control on complex terrains remains a significant challenge, often necessitating expert teleoperation. This stems from the high degree of robot joint freedom need for flipper coordination based terrain roughness. To address this problem, we propose F lipper‐ T rack R obot Bench mark ( FTR‐Bench ), simulator featuring flipper‐track tasked crossing various obstacles using reinforcement learning (RL) algorithms. The primary objective is to enable autonomous locomotion environments that too remote or hazardous humans, such as disaster zones planetary surfaces. Built Isaac Lab, achieves efficient RL training at over 4000 FPS an RTX 3070 GPU. Additionally, it integrates algorithms OpenAI Gym interface specifications, enabling fast secondary development verification. On basis, provides series standardized RL‐based benchmarking experiments baselines obstacle‐crossing tasks, providing solid foundation subsequent algorithm design performance comparison. Experimental results empirically indicate SAC performs relatively well single mixed traversal, but most struggle multi‐terrain traversal skills, which calls community more substantial development. Our project open‐source https://github.com/nubot-nudt/FTR-Benchmark .

Language: Английский

Citations

1
On design of cognitive situation-adaptive autonomous mobile robotic applications DOI Creative Commons
Daniel Pakkala,

Niko Känsäkoski,

Tapio Heikkilä

et al.

Computers in Industry, Journal Year: 2025, Volume and Issue: 167, P. 104263 - 104263

Published: Feb. 18, 2025

Citations

1
AI Vision and Machine Learning for Enhanced Automation in Food Industry: A Systematic Review DOI
D. N. Saha, Mrutyunjay Padhiary, Naveen Chandrakar

et al.

Food and Humanity, Journal Year: 2025, Volume and Issue: unknown, P. 100587 - 100587

Published: March 1, 2025

Language: Английский

Citations

1