RBE PhD. DISSERTATION PROPOSAL
Communication Algorithms for Spatio-Temporal Cooperation in Multi-Robot Systems
Monday, September 28, 2020
1:00 PM - 2:00 PM
Virtual | Email student for zoom link: email@example.com
Abstract: Swarm robotics has the potential to offer key solutions for large-scale, time-sensitive and dangerous applications, such as wildfire fighting and disaster response. Teams of robots promise capabilities beyond the reach of single-robot solutions by distributing intelligence, sensing and actuation at a large scale. This opportunity comes with the challenge of dealing with large amounts of data which are physically distributed across robots. Therefore, a key precondition for the swarm to coordinate successfully is the ability of the robots to store and exchange information efficiently.
In my work, I tackled key aspects of organizing communication in highly mobile robotics swarms. My first step towards this goal was to solve the problem of maintaining global connectivity of the robot swarm, i.e. making sure there are communication paths between any two robots to exchange information. A second step was to realize that, in order to construct a shared database in practice, one must accept that global connectivity might not be possible because of high mobility, obstacles, or challenging mission constraints. Therefore, I designed a distributed data structure for low-memory, low-bandwidth, highly mobile swarms. Thirdly, I developed distributed learning and collective perception applications where the main lever is structured communication enabling data sharing and decision-making throughout robot teams.
The outcomes of this work include (1) scalable connectivity maintenance algorithms tested extensively in realistic simulation and with real robots, (2) a general and reusable platform for storing quantities of data that exceed the memory of individual robots, while maintaining near-perfect data retention in high-load conditions, and (3) algorithms for collectively learning a machine learning model and improving accuracy of predictions through collaboration.
Professor Carlo Pinciroli, Worcester Polytechnic Institute (WPI)
Prof. Giovanni Beltrame, École Polytechnique de Montréal
Prof. Jie Fu, Worcester Polytechnic Institute
Prof. Alexander Wyglinski, Worcester Polytechnic Institute