Giulio Sandini From Active Vision to iCub and the Elusive Quest for Cognition

In this talk

1. 1:10

   From visual processing to active vision

   Sandini traces his work from coordinating systems and neuroscience experiments to studying vision. Early research focused on extracting edges and object shapes, while his space-variant sensor explored how the retina’s uneven layout could reduce energy use. He describes active vision and Rodney Brooks’s work as shifts toward understanding intelligence through action and context.

2. 4:21

   Building robots to study development

   Sandini recalls a European project with David Vernon on understanding images and movement, a theme that continued across his career. Work with researchers including Luciano Fadiga and Claus von Hofsten helped bring cognition into robotics. The RobotCub project aimed to build a robot for studying cognitive development, while his work at IIT connected research in natural and artificial systems with technology and society.

3. 7:30

   Why capable bodies still need imagination

   Sandini says robotics has gained capable bodies and useful insights into neuroscience, but systems still lack a way to anticipate and consider what they might do. He argues that researchers underplayed imagined actions, which can help people devise different routes to a goal and choose among tools. He connects this idea to neuroscience accounts of movement as shaped by internal simulations.

4. 10:27

   A shared architecture for cognition

   Sandini proposes an artificial cognitive architecture that tracks possible actions and selects useful ones. He calls for research that connects brain structure and function, learned data and models, and processes operating across different timescales, from development to evolution. He argues that robotics, neuroscience, and other fields need to work together on concrete projects to build embodied intelligent systems.