Graphical Methods for Multi-Robot Control in Mixed-Competitive Environments
Date and time: 7 November 2024, 13:00-14:00 CET
Speaker: Malintha Fernando, KTH
Title: Graphical Methods for Multi-Robot Control in Mixed-Competitive Environments
Where: Digital Futures hub, Osquars Backe 5, floor 2 at KTH main campus OR Zoom
Directions: https://www.digitalfutures.kth.se/contact/how-to-get-here/
OR
Zoom: https://kth-se.zoom.us/j/69560887455
Host/administrator: TBC
Abstract: As we witness the continuous advancements of mobile robotics with increased autonomous control capabilities, e.g., drones, self-driving taxis and trucks, a naturally rising research question is how to coordinate them effectively in the real-world? This problem is aggravated by the inherentphysical heterogeneity of rapidly developing autonomous systems, i.e., carrying capacity, range, and also by the intentional heterogeneity, i.e., affinity toward a particular group, and the level of competition in the system.
In this talk, we explore a couple of recent trends in multi-agent reinforcement learning (MARL) literature and discuss adapting them in real-world multi-robot systems operating under such constraints: 1) learning policies in mixed-motive settings, and 2) learning scalable policies with graph neural networks.
First, we will discuss multi-robot applications where we cannot design fully cooperative objectives due to real-world constraints by resorting to classical problems such as flocking and coverage. Next, we will start from classical graphical control methods and draw parallels to the current MARL literature, by introducing more stochasticity and dynamicity that often arise in real-world multi-robot literature under communication and heterogeneity.
Bio: Malintha Fernando is a digital futures postdoctoral fellow at the KTH Royal Institute of Technology, and he obtained his Ph.D in 2023 from Indiana University, Bloomington (USA). Malintha’s research focus on designing cooperative control policies for multi-robot systems that operate under limited communication constraints. Prior to joining KTH, Malintha worked as a full-time visiting lecturer in Machine Learning at Indiana University.