Enhancing Human-Robot Interaction by Detecting and Modulating Information Flows

This thesis explores how robots can better understand and respond to subtle social signals when interacting with humans. Using a mathematical tool called Transfer Entropy, it first develops a method to detect and analyse these signals, such as body movements or timing cues in activities like following someone, handing over an object, or joining a group. It then applies this knowledge to train robots, guiding them to influence interactions in helpful ways. Tests in both simulations and real-world experiments show that the approach makes robot behaviour clearer, fairer, and more adaptive, paving the way for smoother and more natural human–robot collaboration.