Using Artificial Neural Networks to Understand How Peripersonal Neurons Map Space Around the Body

Recent research by scientists from the Chinese Academy of Sciences and the Italian Institute of Technology has advanced our understanding of how the brain perceives the space immediately surrounding the body, known as peripersonal space. This area is crucial for interactions with objects and other individuals in our environment. By employing brain-inspired computational models, researchers uncovered that receptive fields surrounding different body parts are organized into a modular system that assists in constructing an internal map of the space near us.

The study, published in Nature Neuroscience, was initiated somewhat unexpectedly during preliminary experiments when researchers observed that the hand-blink reflex—an involuntary blink triggered by hand stimulation—varied depending on the hand’s position relative to the eye. This echoed properties of peripersonal neurons, which respond to stimuli close to the body. Recognizing gaps in existing explanations about these neurons, including their response to stimulus valence and speed, the team aimed to develop a comprehensive quantitative framework.

Instead of solely gathering new biological data, they created artificial neural networks trained with reinforcement learning to simulate how neurons respond during object interception and avoidance tasks. These models learned to evaluate the potential reward or punishment of actions, reflecting how real neurons might process surrounding stimuli. The researchers proposed an "egocentric value map"—a predictive model derived from neural activity patterns—that encapsulates the immediate environment from the perspective of the body.

This computational approach successfully recreated the behavior of biological peripersonal neurons, with the artificial models’ receptive fields expanding in response to faster-moving stimuli, tool use, and high-value objects. The networks naturally developed specialized sub-networks for actions like avoidance and interception, resembling the modularity seen in primate brains. When compared to experimental neural data, the artificial neurons displayed body-part-centered receptive fields that aligned with actual biological observations.

The findings indicate that peripersonal neurons are fundamental in forming an internal, predictive map of the environment tailored to the body's position. The researchers’ framework outperformed other interpretations by fitting extensive empirical data, providing both theoretical insight and practical potential. These results could inform the development of embodied AI, robotic systems, and neuroprosthetics, enabling more adaptive and context-aware interactions.

Looking ahead, the team plans to refine their models by incorporating factors such as sensory uncertainty through advanced mathematical frameworks like active inference. Their goal is to create more nuanced and realistic simulations of neural responses, fostering applications in neuroprosthetics and human-robot interaction that could lead to more natural and effective collaboration.

In summary, this research bridges neuroscience and artificial intelligence, demonstrating how brain-inspired models can elucidate the sophisticated processing underlying our perception of space around us. It opens new pathways for understanding neural mechanisms and designing smarter robots and prosthetic devices that navigate and interact within our shared environment seamlessly.