摘要
当前,传统的智能巡检机器人面临两大潜在挑战:第一,其运行受限于导轨系统,导致巡检覆盖区域有限,每当生产任务或产线布局发生变动时,就必须频繁调整导轨设计,这无疑增加了不必要的成本负担;第二,为提升性能,这类机器人需持续与环境进行交互,这不仅会增加机器人器件损坏的风险,还可能带来潜在的安全隐患。针对上述问题,提出了一种基于Q值正则化的数据增强型离线强化学习算法(Q Regular Term Offline Reinforcement Learning,QROL),其核心思想是利用离线强化学习算法不需要与环境交互的特性,仅利用历史数据集,智能巡检机器人便能自主完成学习过程,摆脱对导轨的依赖,实现灵活的路径规划与自动避障功能。为验证QROL算法的有效性,在动车巡检仿真平台上进行了实验。
Traditional robots face two key challenges:they are limited by guideway systems,which restrict their inspection coverage and require frequent adjustments with production changes,and they need continuous environmental interaction,which risks damaging components and poses safety hazards.To overcome these issues,a data-augmented offline reinforcement learning algorithm(QROL)is proposed in this paper.This algorithm leverages the ability of offline learning to not require environmental interaction,enabling robots to independently complete the learning process using only historical datasets.This allows robots to achieve flexible path planning and automatic obstacle avoidance without relying on guideways.
出处
《工业控制计算机》
2025年第12期65-66,69,共3页
Industrial Control Computer
关键词
智能巡检机器人
离线强化学习
路径规划
自我学习
数据增强
intelligent inspection robot
offline reinforcement learning
path planning
self-learning
data augmentation
