Towards Correct-by-design Social Autonomy
Date and time: 6 May 2021, 12:00 – 13:00 CEST (UTC +2)
Speaker: Alexis Linard, Postdoctoral researcher at Division of Robotics, KTH
Title: Towards Correct-by-design Social Autonomy
Zoom: https://kth-se.zoom.us/j/69560887455
Meeting ID: 695 6088 7455
Password: 755440
Watch the recorded presentation:
Abstract: Deploying robots in social environments leads to new research challenges, such as predicting human behaviour in crowds or identifying different navigation styles. The goal is to enable robots to react efficiently and appropriately to human behaviour, and why not, to mimic it. Methods based on deep neural networks bring powerful solutions: however, they are essentially black boxes making decisions that are hardly decipherable. How can we provide guarantees concerning the validity and the safety of human-robot interaction models? And how can we build accountable and interpretable models for safety-critical human-robot interactions?
In this talk, I will cover several topics, ranging from perceived safety in Reinforcement Learning to modelling human preferences in robot navigation, thanks to formal methods. I will also mention how we can learn formal specifications from data and to what extent temporal logic inference is an interesting first step towards correct-by-design social autonomy.
Bio: Alexis Linard is a Postdoctoral researcher at KTH, Division of Robotics, Perception and Learning. He received an MSc degree in Natural Language Processing from the University of Nantes, France, in 2015, and his PhD from Radboud University Nijmegen, the Netherlands, in 2019. He now works as part of the WASP Expedition Project CorSA: Correct-by-design and Socially Acceptable Autonomy, in collaboration with Iolanda Leite and Jana Tumova. His research focuses on Cyber-Physical Systems, Reinforcement Learning, Temporal Logic Inference and, more broadly, Formal Methods applied to Social Robotics.