This master thesis has been published in 2019.
The safe navigation and execution of tasks, such as assisting humans, in a crowded environment is still a difficult task for mobile robots and is still not fully solved. As a first step towards solving this issue, we present in this thesis an active SLAM algorithm for static environments, which we then extend to crowded environments. The goal of active SLAM is to give a robot the ability to explore an unknown environment, even if it is crowded, in an optimal way. We use a utility function based on Rényi’s general theory of entropy, which automates the trade off between exploration and exploitation without the necessity to manually tune parameters. We show that the free scalar parameter in Rényi’s entropy was not related correctly to the pose uncertainty. To improve this, we present a relation, which scales to the map resolution. We further show that moving people can be detected in 2D range data by using an adaptive breakpoint detector for clustering possible candidates. We then track the clustered objects using individual Kalman Filters. The tracks are then distinguished using counter, size and velocity checks. Further, we use an adaptive people ellipse, which adapts its form depending on the current velocity and moving direction, to merge leg clusters, obtained from a 2D LIDAR, for better tracking of people. Finally, we show results both in a simulation and real environment for static and crowded cases. We can show that our approach achieves better results than a simple shortest frontier strategy and that the detection and tracking of moving people in a crowded environment clearly improves the performance of our active SLAM framework
- Title: Active SLAM in Crowded Environments
- Authors: Dario Mammolo
- Supervisors: Daniel Dugas, Jen Jen Chung, Mark Pfeiffer, Roland Siegwart
- Date of publication: 20/03/2019
- View article