亲爱的研友该休息了!由于当前在线用户较少,发布求助请尽量完整地填写文献信息,科研通机器人24小时在线,伴您度过漫漫科研夜!身体可是革命的本钱,早点休息,好梦!

SPHTRLM: secure and privacy-preserving hyperparameter-tuned reinforcement learning method for robot path finding in dynamic environments

计算机科学 强化学习 路径(计算) 机器人 人工智能 机器人学习 运动规划 移动机器人 控制(管理) 机器学习 增强学习 机器人学 模拟 人机交互 机器人控制 钢筋
作者
Revati Raman Dewangan,Deepali Thombre,Vivek Parganiha,Monika Verma,Amit Purushottam Pimpalkar,Bhupesh Kumar Dewangan,Nilesh M Shelke
出处
期刊:Scientific Reports [Nature Portfolio]
卷期号:16 (1) 被引量:3
标识
DOI:10.1038/s41598-026-48141-x
摘要

Autonomous robot navigation within a dynamic environment is a complicated issue since environmental factors keep on changing, safety remains a factor, and issues of data privacy concern are also on the increase. The existing reinforcement learning (RL) navigation systems mainly focus on path performance and avoidance of collisions but do not focus on privacy protection, adaptation learning stability, and real deployment. This research aims to overcome these constraints by suggesting a novel framework Secure and Privacy-Preserving Hyperparameter-Tuned RL Model (SPHTRLM) to the efficient generation of path plans in grid ecosystems with dynamic environments. The framework incorporates adjusted Q-learning with federated learning (FL) based distributed updates, refined differentiated privacy, minimal encrypted parameter exchange, adaptive reward shaping and automatic hyperparameter optimization. In a further attempt to enhance practicability, the proposed architecture also embraces mobility conscious aggregation and heterogeneous model support of resource-limited robotic platforms. The suggested SPHTRLM has a success rate of (95% ± 2%), and it is better than the comparable one Q-learning (87% ± 4%) and Deep RL (DRL) baselines (88%) when these methods were evaluated under the same condition. The framework minimizes distances to the average path with a reduction of 20–25% and convergence is speeded up by around 35% compared to normal Q-learning. When the obstacles are very thick then the collision rate becomes and the obstacle reduces to 0.08, and the safety of the navigation process improves. Although there are additional privatization mechanisms, the computational costs are minimal (8–12%), and the average decision time is 110–125 ms, which meets the real-time operational capabilities. Privacy analysis with formally stated membership inference and reconstruction attacks provide status of attack rate less than 5% attack success with both white and black box adversary. These findings underscore that SPHTRLM is a feasible way of achieving the goals of ensuring navigation, learning consistency, safety as well as privacy protection to give credible acceptance to using autonomous robotic systems in dynamic and data-sensitive environment.

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
神勇千秋完成签到,获得积分10
1秒前
沈惠映完成签到 ,获得积分0
6秒前
7秒前
FashionBoy的应助被iman采纳,获得10
10秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
10秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
10秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
10秒前
乐乐的应助被纯情的阁采纳,获得10
11秒前
h11的应助被纯情的阁采纳,获得10
11秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
11秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
11秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
11秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
11秒前
12秒前
大气钧完成签到,获得积分10
15秒前
传奇3的应助被纯情的阁采纳,获得20
20秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
20秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
20秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
20秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
20秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
20秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
21秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
21秒前
madeleinli的应助被纯情的阁采纳,获得10
21秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
21秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
21秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
21秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
21秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
21秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
21秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
22秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
22秒前
科研通AI6.2的应助被纯情的阁采纳,获得10
22秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
22秒前
九霄完成签到 ,获得积分10
22秒前
科研通AI6.4的应助被纯情的阁采纳,获得10
22秒前
24秒前
完美灵松发布了新的文献求助10
31秒前
32秒前
36秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1500
The World Atlas of Wine 9th edition 1000
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
John McDowells Bildungstheorie 800
八旗制度の研究 600
Analyses de reconnaissances de dettes passées devant les échevins d'Ypres (1249-1291) 600
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8158848
求助须知:如何正确求助?哪些是违规求助? 9554492
关于积分的说明 21719333
捐赠科研通 7748419
什么是DOI,文献DOI怎么找? 3427099
关于科研通互助平台的介绍 2552006
邀请新用户注册赠送积分活动 2471370