Research Vision

I admire the complexity of the universe and enjoy the knowledge that we have acquired and structured as a species. I am fascinated by how plants, animals and humans, both individually and at the group level, adapt to and shape their environment. Decades of research in biology, artificial intelligence and robotics have uncovered some of the mechanisms underlying these capabilities in specific contexts. Yet, as demonstrated by the very limited flexibility of existing robots and video game agents, how to extend these to new situations is an open research question. Therefore, my long-term research goal is to find task-independent mechanisms of autonomy and adaptation.

I pursue this goal by taking an experimental approach; after having worked on robotics models of living systems and robots shaping their environments (video), my recent research has focused on developing methods to allow humans to help robots and intelligent agents to become autonomous. Indeed, I believe that by studying how autonomy can be imparted on agents in various situations, we can understand its fundamental properties. I have approached this target from two complementary directions.

The first aims at enabling beginners – including young children – to program robots graphically (video). Programming a robot requires to make it able to take action by itself given its inputs, and studying that helps understanding the representation of autonomous behaviours, but also the nature of human learning.

The second direction is to construct models to teach behaviours to robots. It studies the behaviour representation, and explores how the robot can learn and generalise a behaviour from a set of demonstrations by a human.

These two research orientations are dual and both are centred around the learning process: the human learns from interacting with the robot in the first and the robot learns from the guidance of the human in the second. Because they cover the same fundamental questions of generalisation, abstraction, symbol creation, etc. they form a unique opportunity to study the fundamental building blocks of an autonomous and adaptive behaviour.

Finally, being applied to real-world problems, my approach to research provides innovative solutions for education and agents programming. For instance, the work on graphical programming is deployed on the Thymio robot (video), of which 30 k units have been sold to date.

© 2008–2022 Stéphane Magnenat
This page is valid XHTML 1.0 Transitional and CSS 3.0 with text-rendering extension.
Privacy policy