Development of a Duct Cleaning Robot Algorithm Using Reinforcement Learning-based Autonomous Driving

Dust accumulation in indoor air purification built-in ducts is inevitable. To reduce air pollution and the risk of fire, frequent dust removal is essential. However, conventional duct cleaning methods are time- and cost-extensive and cleaning results are not satisfactory. To resolve this issue, we previously suggested a quantitative dust determination algorithm using machine learning. However, as autonomous driving is equally essential, in this study, reinforcement learning is applied to autonomous driving of the cleaning robot, based on which a cleaning robot algorithm is developed. In a 3D virtual environment, reinforced (driving) training was conducted for various duct shapes. Through this process, collision with the duct wall can be avoided and the optimum route can be obtained. The dust amount is determined during drive by the duct cleaning robot which uses a previously developed dust quantification algorithm. The algorithm presented in this study is validated by installing it on the robot. It can be confirmed that reinforcement learning-based self-driving together with the previously suggested dust quantification algorithm is effective and is applicable to autonomous driving.