A Decision-Making Algorithm of Multiple Reactive Tasks for Autonomous Driving Sweepers Based on Behavior Trees

Autonomous driving sweepers typically need to manage multiple reactive tasks, such as battery and trash level monitoring, requiring immediate action when conditions are met. Therefore, equipping an intelligent decision-making algorithm is undoubtedly necessary for ensuring their efficiency and high accuracy of executing special operations. Traditional rule-based methods are often insufficient for timely responses to these tasks, and current learning-based approaches, while performant, can lack absolute correctness in all scenarios. Consequently, in this article, a novel decision-making method designed to meet the demands of multiple reactive tasks is proposed to serve autonomous driving sweepers. Smart nodes (smart sequence and smart fallback) are introduced to the behavior trees framework, selectively executing child nodes with the “smart” attribute for real-time task checking. In addition, a robust checkpoint recovery mechanism ensures the system returns seamlessly to the previously interrupted node after a reactive task is completed. Building on this, a method for constructing behavior trees from smart nodes translates decision tasks from natural language into practical applications. Experiments on an autonomous sweeper demonstrate the effectiveness of this approach in managing multiple reactive tasks.