AI for autonomous robotics

In computer science, machine learning has defined a set of proven statistical techniques that can to some extent be compared with forms of learning in living organisms. However, their implementation in autonomous robotics highlights a number of weaknesses in ensuring agent autonomy. The aim of this course is to revisit these techniques in the light of data from the neurosciences and social sciences, and to present algorithms that enable autonomous learning through simple interaction with the environment, with survival criteria defined a priori. For each form of learning, after a reminder of the classical forms of automatic learning, autonomy criteria are defined and biological and behavioral data are introduced, enabling more biologically plausible forms to be defined, integrating a more global systemic view of living organisms.

1. Principles of learning and autonomy in the living world
2. Social and imitative learning
3. Unsupervised and supervised learning
4. Intrinsic motivation and curiosity
5. Motivated learning