Controlling robots for high-Level scene understanding (Eduardo Montijiano) ; Spatiotemporal logic control for heterogeneous multi-agent systems (Dimos Dimarogonas)

Starting on
IRISA Rennes
Eduardo Montijiano and Dimos Dimarogonas

For internal staff only

Eduardo Montijiano 

Title: Controlling robots for high-Level scene understanding 

Abstract: The problem of automatically interpreting the surroundings of intelligent systems, scene understanding, is a widely studied problem that includes all kinds of recognition tasks from different sensor data. Current state of the art in machine learning includes impressive results for several scene understanding tasks, but, to be used in real robots, these methods still do not offer the required levels of interpretability that classic sensing algorithms provide. In this talk, I will explain how we can expand the opportunities to apply scene understanding methods to more complex scenarios by adding mobility to the sensors, describing new measurement models, fusion algorithms and planning techniques tailored to be used onboard mobile robots that perceive the world with the latest machine learning tools. 

BIO: Eduardo Montijano is an Associate Professor in the Departamento de Informatica e Ingenieria de Sistemas at Universidad de Zaragoza in Spain. He received the M.Sc. and Ph.D. degrees from the Universidad de Zaragoza, Spain, in 2008 and 2012 respectively. He was a faculty member at Centro Universitario de la Defensa, Zaragoza, between 2012 and 2016. His research interests are in the field of distributed algorithms applied to cooperative control and perception of multi-robot systems. 


Dimos Dimarogonas 

Title: Spatiotemporal logic control for heterogeneous multi-agent systems 

Abstract: Formal methods-based planning and control is an established methodology for autonomous systems that are subject to high level tasks. When it comes to multi-agent systems, recent trends involve objectives which are quantified in space and time in the form if Signal Temporal Logic (STL) formulas and its variations. However, in most cases the agents are considered homogeneous with respect to their dynamics, sensing and communication constraints and actuation limitations. In this talk, we consider the problem of control synthesis for heterogeneous multi-agent systems under STL tasks, including the case of leader-follower networks where an agent subgroup of leaders are responsible for the control objective fulfilment of the whole group. We start by presenting transient controllers for heterogeneous networks and under simple control objectives such as consensus and formation control, and we then use these controllers as bases for more general multi-agent tasks given by STL. We finally consider the potential discrepancies between the task dependency and communication graphs, and propose the means on how to account for such discrepancy in the control design. 

BIO: Dimos Dimarogonas is Full Professor in Control and Robotics at the School of Electrical Engineering and Computer Science, KTH Royal Institute of Technology, Stockholm, Sweden. He graduated from the School of Electrical and Computer Engineering of the National Technical University of Athens (NTUA) in 2001 and was awarded a PhD in Mechanical Engineering at NTUA in 2007. He held postdoctoral positions at KTH and MIT between 2007 and 2010 and has joined KTH as an Assistant Professor in 2010. He is currently serving as Professor of Automatic Control in the Division of Decision and Control Systems, EECS, and currently serving as Associate Editor for Automatica and the IEEE Transactions on Control of Network Systems.