Informed expansion for informative path planning via online distribution learning

• Published in: Robotics and autonomous systems 2023-08, Vol.166, p.104449, Article 104449
• Main Authors: Zacchini, Leonardo, Ridolfi, Alessandro, Allotta, Benedetto
• Format: Article
• Language: English
• Subjects: Autonomous Underwater Vehicles; Informative Path Planning; Informed expansion; Mobile robots
• ISSN: 0921-8890; 1872-793X
• DOI: 10.1016/j.robot.2023.104449

Mobile robots are essential tools for gathering knowledge of the environment and monitoring areas of interest as well as industrial assets. Informative Path Planning methodologies have been successfully applied making robots able to autonomously acquire information and explore unknown surroundings. Rapidly-exploring Information Gathering approaches have been validated in real-world applications, proving they are the way to go when aiming for Information Gathering tasks. In fact, RIG can plan paths for robots with several degrees of freedom and rapidly explore complex workspaces by using the state-of-the-art Voronoi-biased expansion. Nevertheless, it is an efficient solution when most of the area is unknown but its effectiveness decreases as the exploration/gathering evolves. This paper introduces an innovative informed expansion for IG tasks that combines the Kernel Density Estimation technique and a rejection sampling algorithm. By learning online the distribution of the acquired information (i.e., the discovered map), the proposed methodology generates samples in the unexplored regions of the workspace, and thus steers the tree toward the most promising areas. Realistic simulations and an experimental campaign, conducted in the underwater robotics domain, provide a proof-of-concept validation for the developed informed expansion methodology and demonstrate that it enhances the performance of the RIG algorithm. •Informed sampling methodology for enhancing sampling-based IG algorithm performance.•Informed expansion for biasing RIG expansion toward unexplored areas.•KDE technique for learning online the distribution of the acquired data.•Single-stage planning paradigm for exploration tasks that can run on compact robots.•Real-world experiments conducted in the context of underwater robotics.