A revolutionary breakthrough in social robotics is changing how machines learn to interact with humans. Researchers have developed a simulation system that allows social robots to be trained without requiring human participants, potentially transforming the field's development timeline.
The study, presented at the 2025 IEEE International Conference on Robotics and Automation (ICRA), was conducted by a team from the University of Surrey and the University of Hamburg. Their approach centers on a dynamic scanpath prediction model that helps robots anticipate where humans would naturally look during social interactions.
"Our method allows us to test whether a robot is paying attention to the right things – just as a human would – without needing real-time human supervision," explains Dr. Di Fu, co-lead of the study and lecturer in Cognitive Neuroscience at the University of Surrey.
The research team validated their model using two publicly available datasets, demonstrating that humanoid robots could successfully mimic human-like eye movements. By projecting human gaze priority maps onto a screen, they directly compared the robot's predicted attention focus with real-world data, eliminating the need for large-scale human-robot interaction studies in early research phases.
This innovation addresses a significant bottleneck in social robotics development. Previously, researchers needed numerous human participants to train and test robots designed for social settings like education, healthcare, and customer service. Examples of such robots include Pepper, a retail assistant, and Paro, a therapeutic robot for dementia patients.
By enabling researchers to test and refine social interaction models at scale through simulation before real-world deployment, this breakthrough could dramatically accelerate the development cycle of social robots while reducing costs and improving their effectiveness in human environments.