Development of a portable, multi-agent experimental platform for demonstration of a persistent coverage control algorithm

Development of a portable, multi-agent experimental platform for demonstration of a persistent coverage control algorithm

Vermeulen, Arjan (TU Delft Mechanical, Maritime and Materials Engineering)

van den Boom, Ton (mentor)
Keviczky, Tamas (mentor)
Mendel, Max (graduation committee)

Delft University of Technology

Mechanical Engineering | Systems and Control

2019-12-19

At DCSC, research is conducted in the field of multi-agent systems. While this research often takes a theoretical approach, there is a growing need to verify the research on an experimental platform. In this way, the newly developed methods can be experimentally validated. For this purpose, a compact system is developed for localization and pose estimation of small, differential drive agents, using a camera and computer vision. This platform is designed in such a way that it is both portable and quick to set up (< 5 minutes). This platform is developed for two main purposes: verification of developed control methods for multi-agent systems on a physical system, and for the demonstration of various control methods on location, e.g. in a classroom. The requirements for such a platform are formulated and the components needed to fulfill these requirements are selected. The performance of the localization and pose estimation in terms of computational time, precision and accuracy is quantified. To be able to deploy the setup on a table top of various sizes, multiple methods are designed and implemented to keep all the agents within a predefined domain. A Python based program is developed for demonstration and validation of control strategies. A graphical user interface is developed for the demonstration of path-planning and control algorithms, providing feedback on the states of the individual agents, and provide for real-time switching of control objective, observer model and visualizations. This platform is developed, built and tested. Coverage control aims to direct the motion of multiple agents/robots in a way to cover an area. In persistent coverage problems, points in the area to be covered need to be revisited regularly, since their time-varying "coverage level", diminishes over time. The coverage level could represent a variety of real-world quantities. Typical examples include tasks such as lawn mowing, visual inspection or robotic cleaning tasks. A state-of-the-art persistent coverage control algorithm for non-holonomic agents is implemented on the experimental platform, and the results are compared with simulations, showing a similar evolution of the coverage level, with a slightly higher variance in the coverage level in practice versus simulations. Furthermore, improvements to the partitioning scheme are proposed, tested and implemented. This novel partitioning achieves marginally higher coverage levels for agents with a constant coverage production function.

Persistent Coverage
Nonholonomic Motion Planning
Experimental platform

http://resolver.tudelft.nl/uuid:84e4e2aa-464c-4629-bbf9-6be2e07f3412

Student theses

master thesis

© 2019 Arjan Vermeulen