Alessandra Sciutti (Istituto Italiano di Tecnologia) - Cognitive robotics for collaboration

In this talk

1. 1:30

   Why social understanding is hard for robots

   Sciutti contrasts robots’ growing physical and strategic abilities with their difficulty reading ordinary social signals, such as an offer to shake hands. Human perception relies on experience and expectations, which can make it fast and useful while also introducing biases. Movement style also carries meaning: people can distinguish a gentle object handoff from an aggressive one even without seeing the actor’s face.

2. 15:39

   Movement can shape collaboration

   In experiments, people unconsciously adjusted their walking speed when a robot changed pace, but this effect depended on the robot using humanlike acceleration and deceleration. When researchers adapted the full motion pattern of gentle or rude handoffs, people judged the robot’s style similarly to a human’s, and their own receiving movements changed to match. A study in Italy and Japan also found speed matching in a cooperative task in both locations.

3. 28:44

   Robots can read motion and gaze

   Sciutti describes a robot that learned to distinguish biological from mechanical movement using optical flow and simple motion features, without needing to detect a human skeleton or skin. It could orient toward a person moving an object even when that person was out of sight. In a separate block-passing task, gaze alone helped the robot choose what to hand over and when; wearable eye tracking also detected signs of increased cognitive load associated with lying, with 73% accuracy in one task.

4. 23:51

   Interaction changes perception and trust

   When participants judged distances alongside a socially behaving robot, they relied less on the history of previous distances than when paired with a mechanical-looking robot. Sciutti says this suggests social interaction can shift basic perceptual strategies. In a 30-minute treasure hunt, participants increasingly sought the robot’s help, and those who lost after following its gamble suggestion reported the largest increase in trust.

5. 44:05

   Building toward adaptable social partners

   Sciutti says isolated signals are only early components of sustained interaction, where people and robots need to adapt to each other over time. She points to the need for systems that combine perception, action, memory, and prediction, and for better ways to adapt to another person’s expectations. In discussion, she identifies timing, turn-taking, memory, and selecting useful experience as open technical challenges.