UAV Collision Avoidance in Unknown Scenarios with Causal Representation Disentanglement DOI Creative Commons
Zhun Fan,

Zihao Xia,

Che Lin

et al.

Drones, Journal Year: 2024, Volume and Issue: 9(1), P. 10 - 10

Published: Dec. 25, 2024

Deep reinforcement learning (DRL) has significantly advanced online path planning for unmanned aerial vehicles (UAVs). Nonetheless, DRL-based methods often encounter reduced performance when dealing with unfamiliar scenarios. This decline is mainly caused by the presence of non-causal and domain-specific elements within visual representations, which negatively impact policies. Present techniques generally depend on predefined augmentation or regularization intended to direct model toward identifying causal domain-invariant components, thereby enhancing model’s ability generalize. However, these manually crafted approaches are intrinsically constrained in their coverage do not consider entire spectrum possible scenarios, resulting less effective new environments. Unlike prior studies, this work prioritizes representation presents a novel method disentanglement. The approach successfully distinguishes between data. By concentrating aspects during policy phase, factors minimized, improving generalizability DRL models. Experimental results demonstrate that our technique achieves reliable navigation collision avoidance unseen surpassing state-of-the-art deep algorithms.

Language: Английский

Anytime algorithm based on adaptive variable-step-size mechanism for path planning of UAVs DOI Creative Commons
Hui Gao,

Yuhong Jia,

Liwen XU

et al.

Chinese Journal of Aeronautics, Journal Year: 2024, Volume and Issue: unknown

Published: Sept. 1, 2024

Language: Английский

Citations

1
Hybrid Machine Learning and Reinforcement Learning Framework for Adaptive UAV Obstacle Avoidance DOI Creative Commons
Wojciech Skarka, Rukhseena Ashfaq

Aerospace, Journal Year: 2024, Volume and Issue: 11(11), P. 870 - 870

Published: Oct. 24, 2024

This review explores the integration of machine learning (ML) and reinforcement (RL) techniques in enhancing navigation obstacle avoidance capabilities Unmanned Aerial Vehicles (UAVs). Various RL algorithms are assessed for their effectiveness teaching UAVs autonomous navigation, with a focus on state representation from UAV sensors real-time environmental interaction. The identifies strengths limitations current methodologies highlights gaps literature, proposing future research directions to advance technology. Interdisciplinary approaches combining robotics, AI, aeronautics suggested improve performance complex environments.

Language: Английский

Citations

1
A Novel Path Planning Approach Based on Exploration-efficient Reinforcement Learning DOI
Meiying Cai, Yong Tang,

Yanwei Du

et al.

2022 34th Chinese Control and Decision Conference (CCDC), Journal Year: 2024, Volume and Issue: unknown, P. 385 - 389

Published: May 25, 2024

Language: Английский

Citations

0
Deep reinforcement learning based integrated evasion and impact hierarchical intelligent policy of exo-atmospheric vehicles DOI Creative Commons

Leliang Ren,

Weilin Guo, Yong Xian

et al.

Chinese Journal of Aeronautics, Journal Year: 2024, Volume and Issue: 38(1), P. 103193 - 103193

Published: Aug. 24, 2024

Language: Английский

Citations

0
Cloud-based Speech Recognition for UAV Control Architecture in Industry 4.0 DOI

Oghenegueke P Eruero,

Modestus O. Okwu,

Favour O. Eric

et al.

Published: Aug. 1, 2024

Language: Английский

Citations

0
Autonomous road hazard detection and avoidance system using deep reinforcement learning for intelligent vehicles DOI

S Swathi,

Saranya Vinayagam,

J. S. Sujin

et al.

E-Learning and Digital Media, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 17, 2024

Road hazards significantly contribute to fatalities in traffic accidents. As the number of vehicles on road increases, risk accidents rises, especially under adverse weather conditions that impair visibility and conditions. In such scenarios, it is crucial alert approaching prevent further collisions. Detecting humans or animals essential minimize Accurate detection estimation are vital for ensuring safety enhancing driving experience. Current deep learning methods condition monitoring often time-consuming, costly, inefficient, labor-intensive, require frequent updates. Therefore, there pressing need more flexible, cost-effective, efficient process detect conditions, particularly hazards. this work, we present a hazard avoidance system autonomous using reinforcement (DRL) address congestion issues complex We utilize GoogLeNet feature extraction, which extracts features from given images. Subsequently, design modified compact snake optimization (MCSO) algorithm optimization, addressing data dimensionality issues. Additionally, introduce geometric (GDRL) tracking environments, improving accuracy robustness visual detection. The proposed MCSO + GDRL model validated self-made open access dataset with 5607 samples car recorders KITTI training.

Language: Английский

Citations

0
Evade Unknown Pursuer via Pursuit Strategy Identification and Model Reference Policy Adaptation (MRPA) Algorithm DOI Creative Commons

Z. D. Su,

Shuang Zheng, Zhiqiang Xu

et al.

Drones, Journal Year: 2024, Volume and Issue: 8(11), P. 655 - 655

Published: Nov. 8, 2024

The game of pursuit–evasion has always been a popular research subject in the field Unmanned Aerial Vehicles (UAVs). Current evasion decision making based on reinforcement learning is generally trained only for specific pursuers, and it limited performance evading unknown pursuers exhibits poor generalizability. To enhance ability an policy learned by (RL) to evade this paper proposes pursuit UAV attitude estimation strategy identification method Model Reference Policy Adaptation (MRPA) algorithm. Firstly, constructs Markov model UAVs that includes pursuer’s trains using Soft Actor–Critic (SAC) Secondly, establishes novel relative motion games under assumption proportional guidance used as strategy, which algorithm proposed provide adequate information adaptation. Furthermore, presented improve generalizability RL certain environments. Finally, various numerical simulations imply precision accuracy identification. Also, ablation experiment verifies MRPA can effectively deal with pursuers.

Language: Английский

Citations

0
Multi‐target cognitive electronic reconnaissance for unmanned aerial vehicles based on scene reconstruction DOI Creative Commons
Zhang Yun, Shixun You, Yunbin Yan

et al.

IET Radar Sonar & Navigation, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 6, 2024

Abstract Model‐free deep reinforcement learning (DRL) is regarded as an effective approach for multi‐target cognitive electronic reconnaissance (MCER) missions. However, DRL networks with poor generalisation can significantly reduce mission completion rates when parameters such area size, target number, and platform speed vary slightly. To address this issue, paper introduces a novel scene reconstruction method MCER missions group adaptive transfer (MTDRL) algorithm. The algorithm enables quick adaptation of strategies varied scenes by transferring strategy templates compressing perception states. validate the method, authors developed model unmanned aerial vehicle (UAV) MCER. Three sets experiments are conducted varying speed. results show that MTDRL outperforms two commonly used algorithms, 18% increase in rate 5.49 h reduction training time. Furthermore, much higher than typical non‐DRL UAV demonstrates stable hovering repeat behaviours at radar detection boundary, ensuring flight safety during

Language: Английский

Citations

0
Driving risks from light pollution: an improved YOLOv8 detection network for high beam vehicle image recognition DOI
Lili Zhang, Ke Zhang, Kang Yang

et al.

The Journal of Supercomputing, Journal Year: 2024, Volume and Issue: 81(1)

Published: Dec. 11, 2024

Language: Английский

Citations

0
Vision-Based Deep Reinforcement Learning of Unmanned Aerial Vehicle (UAV) Autonomous Navigation Using Privileged Information DOI Creative Commons

Junqiao Wang,

Zhongliang Yu, Dong Zhou

et al.

Drones, Journal Year: 2024, Volume and Issue: 8(12), P. 782 - 782

Published: Dec. 22, 2024

The capability of UAVs for efficient autonomous navigation and obstacle avoidance in complex unknown environments is critical applications agricultural irrigation, disaster relief logistics. In this paper, we propose the DPRL (Distributed Privileged Reinforcement Learning) algorithm, an end-to-end policy designed to address challenge high-speed UAV under partially observable environmental conditions. Our approach combines deep reinforcement learning with privileged overcome impact observation data corruption caused by partial observability. We leverage asymmetric Actor–Critic architecture provide agent information during training, which enhances model’s perceptual capabilities. Additionally, present a multi-agent exploration strategy across diverse accelerate experience collection, turn expedites model convergence. conducted extensive simulations various scenarios, benchmarking our algorithm against state-of-the-art algorithms. results consistently demonstrate superior performance terms flight efficiency, robustness overall success rate.

Language: Английский

Citations

0