Dimos Dimarogonas is developing control systems for robots programmed not only to perform a specific task, but also to work with each other. Service robots for offices, factories, and laboratories are potential future applications.
Associate Professor of Automatic Control
Wallenberg Academy Fellow 2015
KTH Royal Institute of Technology
Developing decentralized multi-agent systems, in which robots are able to detect and co-operate with each other.
Two drones with gripping arms are flying around above a test area in the KTH Royal Institute of Technology Smart Mobility Lab. The area is enclosed by netting and cushioned with mattresses. Outside the netting, two doctoral students from Dimarogonas' research team are standing in their stocking feet with remote controls, ready to intervene and retrieve the drones should they collide or get stuck somewhere.
“To work together, robots have to learn to sense their surroundings and take flexible decisions based on the information they gather,” Dimarogonas says.
Systems in which robots and other machines cooperate – known as multi-agent systems – can be used in office environments and laboratories, for example, and also in the transportation sector, in industrial processes or to assist in search and rescue missions. But these applications are still some way away. Dimarogonas explains:
“I’m sure we’ll see more and better service robots in the coming years; we already have automatic vacuum cleaners and lawn mowers, for example. More futuristic robots that are capable of co-operating with each other may not come onto the market as soon as people hope. Further technological developments are needed to ensure they are safe and sophisticated enough. But the technology is on the way, and here in Sweden great progress is being made, at universities and in the industrial sector.”
Knowledge is needed from several scientific fields to develop airborne or terrestrial robots that can interact with each other, and with people – safely and flexibly. Dimarogonas has put together a team with expertise in mathematics, electronics and mechanics. At the moment he is the principal supervisor for eight doctoral students, of whom one is an industry-sponsored German PhD student from Bosch in Stuttgart.
Dimarogonas has always been interested in applications, and is keen to link his research to “the real world”.
“It’s fun to develop theories, carry out experiments, and write scientific articles, but I also think we need to see a practical use for scientific discoveries.”
Dimarogonas was born and grew up in downtown Athens. He studied electronics at the National Technical University of Athens, going on to gain a PhD in automatic control systems and multi-robot applications. In 2007 he moved to Sweden, having been offered a postdoc position with Professor Karl Henrik Johansson at KTH. He also spent a couple of years as a postdoc at MIT in the U.S. Since 2010 he has been back at KTH, where he is currently drawing much inspiration from computer science.
“I needed a new challenge, and saw that computer science offered tools that can be used to ask control systems to perform a sequence of tasks, rather than just single ones. The field is known as ‘formal validation tools for distributed control’, and has been the focus of my research over the past year.”
Much of Dimarogonas’ work entails putting together three fundamental jigsaw pieces: the network component, i.e. how the robots communicate and sense each other; the physical or analogue component; and the software component, i.e. how to give specific digital instructions to the control system.
“We call it ‘hybrid control’. The tricky part is the fusion between the three elements, and studying whether that fusion creates fundamental limitations.”
“The grant gives me academic freedom, and a real boost to my research. In addition to the funding of the project itself, the award is also recognition of my efforts and progress to date in this field.”
With the help of funding from the Knut and Alice Wallenberg Foundation, Dimarogonas wants to develop a decentralized control system, in which each agent constitutes an individual unit, making decisions autonomously. He is using validation to define suitable sequences for advanced tasks that the robots can perform alone and together.
“This project involves the study of limitations in relation to time and probability aspects. In practice, this means, for instance, that we set limits on how many minutes the robots are to surveil an area. Or, if we say there should be a 95-percent probability of a given action succeeding, we need to understand what control mechanisms are needed to achieve this.”
Another dimension is sensing itself – how the robots register their surroundings to avoid colliding or doing the wrong thing. There are limitations on how much information the agents in a multi-agent system can process.
“The question is how to model this, and how it impacts the behavior and results of the system. If you have service robots performing various tasks in an office environment, their interaction with people and other robots needs to be coordinated and optimized. Each robot needs to have reactive abilities, such as collision avoidance behavior.”
Another key factor that Dimarogonas is studying is how robot systems are to become socially acceptable.
“This is all about refining human-robot interaction. Developments in robotics have many dimensions to take into account, which is one reason I really enjoy researching in this field. There are so many challenges and interesting areas to explore.”
Text Susanne Rosén
Translation Maxwell Arding
Photo Magnus Bergström