Èric Pairet Artau
Real-world robots are becoming a vital ingredient in society. Not only they are required to live in environments originally designed for humans but also to perform human tasks. In this context, systems with dual-arm manipulation capabilities come in handy. Given their complexity, it is reasonable to expect robots to be able to learn from humans so they become accessible to non-roboticists-experts. Particularly, a system can learn a set of primitive actions which allows it to overcome unfamiliar scenarios in a human-like manner. To endow a robot with such competences, the system has to (i) learn the dynamics or policy underlying a task or primitive behaviour, and (ii) be able to exploit this knowledge in front of new scenarios. Jointly dealing with the aforementioned challenges constitute the main motivation of this research project.