Perception Is All You Need: A Neuroscience Framework for Low Cost Sensorless Gaze in HRI

A new research paper by Mason Kadem, titled 'Perception Is All You Need: A Neuroscience Framework for Low Cost Sensorless Gaze in HRI,' presents a radical shift in designing gaze-following robots. Instead of relying on expensive sensor arrays (costing $30,000+), complex computer vision algorithms, and raising privacy concerns, the framework exploits the human visual system's own perceptual machinery. It is grounded in three key neuroscience principles: the brain's distributed face-processing network, a high-precision 'convexity prior' that makes us see concave shapes as convex, and a predictive processing hierarchy where top-down expectations override bottom-up sensory data.

This convergence explains why a simple, concave cardboard eye socket with a painted pupil can create the powerful illusion of mutual gaze from virtually any viewing angle. The viewer's own brain effectively completes the computation, acting as the robot's 'actuator.' Kadem has translated this into a practical, open-template robot design with parameterized, interchangeable eye inserts that cost less than a dollar to produce, require no power, and collect no data. The paper also maps out boundary conditions—developmental, clinical, and geometric—predicting where the illusion will succeed or fail.

If successfully leveraged, this perceptual hack could democratize two decades of human-robot interaction research demonstrating that gaze-following improves attention, recall, and learning, particularly in educational or therapeutic child-robot interactions. By removing the massive cost and technical barriers, these benefits could be delivered at a population scale previously unimaginable, making socially assistive robotics accessible worldwide.