Can a robot learn like a child?

Spontaneously acquiring new skills and behaviors: not only humans and animals can do it. Developmental Robotics research is currently working on the creation of truly intelligent robots that can learn in complete autonomy. At ISTC the Laboratory of Computational Embodied Neuroscience (LOCEN) aims to build robots that, just like children, can acquire increasingly complex behaviors based on curiosity and the pleasure to learn. 

Traditional robotic approaches directly program a set of specific skills in robots, so that they can respond to external stimuli in a rather rigid way, without any capacity to flexibly readapt to new conditions and goals. Building robots that can autonomously learn new behaviors and re-adapt to new conditions is the new frontier of Developmental Robotics research.

Within this framework,ai ISTC  the Laboratory of Computational Embodied Neuroscience (LOCEN) aims to understand how various structures of brain, which have emerged during evolution, can allow higher organisms to autonomously develop new skills through the interaction with the environment. The core idea is that this is possible thanks to intrinsic motivations and a hierarchical organization of action. Intrinsic motivations are processes, such as curiosity and the pleasure to learn, that for example guide learning of children at play.

Playing activities might indeed allow robots to acquire many skills that can be later exploited to pursue goals useful for the users. Hierarchical controllers allow robots to learn new skills based on already acquired skills so to acquire increasingly complex behaviors in a cumulative fashion. This goal has a great importance for science, as learning based on intrinsic motivations is a hallmark of the intelligence of humans and other primates.

Moreover, it has huge technological potential: imagine if you can buy a robot, leave it to have its autonomous experience in your kitchen or in a mine, and then come back one month later and ask "Now wash the dishes and put everything in order for me" or "Dig the coal out of the ground and collect it into that container". Autonomous robots could thus be exploited to perform complex tasks and conduct missions in hostile environments. 

Contacts: Gianluca Baldassarre, Marco Mirolli

ISTC Group: Laboratory of Computational Embodied Neuroscience

Relevant Publications

Santucci V. G., Baldassarre G., Mirolli M. (2010). Biological cumulative learning through intrinsic motivations: a simulated robotic study on development of visually-guided reaching. In Proceedings of the Tenth International Conference on Epigenetic Robotics (EpiRob2010), pp.121-128. Lund: Lund University.

Caligiore D., Mirolli M., Parisi D., Baldassarre G. (2010). A bioinspired hierarchical reinforcement learning architecture for modeling learning of multiple skills with continuous state and actions. In Proceedings of the Tenth International Conference on Epigenetic Robotics (EpiRob2010), pp. 27-34.  Lund: Lund University.