Artificial neural networks reveal how peripersonal neurons represent the space around the body

The brains of humans and other primates are known to execute various sophisticated functions, one of which is the representation of the space immediately surrounding the body. This area, also sometimes referred to as “peripersonal space,” is where most interactions between people and their surrounding environment typically take place.

Researchers at Chinese Academy of Sciences, Italian Institute of Technology (IIT) and other institutes recently investigated the neural processes through which the brain represents the area around the body, using brain-inspired computational models. Their findings, published in Nature Neuroscience, suggest that receptive fields surrounding different parts of the body contribute to building a modular model of the space immediately surrounding a person or artificial intelligence (AI) agent.

“Our journey into this field began truly serendipitously, during unfunded experiments done purely out of curiosity,” Giandomenico Iannetti, senior author of the paper, told Medical Xpress. “We discovered that the hand-blink reflex, which is evoked by electrically shocking the hand, was strongly modulated by the position of the hand with respect to the eye.

“We soon realized that this blink reflex behaved much like so-called peripersonal neurons, which are neurons that respond to objects near the body. As we got more familiar with the literature on this type of neuron, however, we noticed that the existing theoretical explanations of their activity fail to explain quite a lot of their properties, such as their modulation by stimulus valence, speed, and motor repertoire.”

Rather than collecting new data, which could then be added to the extensive and disjointed data collected during previous studies, Iannetti and his colleagues set out to develop a new quantitative framework that clarifies why the peripersonal neurons observed in earlier experiments exist and how they work. This framework could then be integrated with existing neuroscience theories.

To develop their framework, they employed artificial neural networks (ANNs) trained via reinforcement learning. These are brain-inspired computational models that can learn to complete various tasks with good accuracy, emulating the connections between neurons.

“In simple terms, we built computer simulations of simplified ‘animals,’ which learned through trial and error to choose actions based on how much reward or punishment those actions would bring over time,” explained Rory John Bufacchi, first author of the paper.

“Our approach involved three main steps. First, our key insight was that peripersonal responses might simply reflect the value of potential actions: whether reaching out to or dodging environmental objects would lead to rewards or punishments.”

Iannetti, Bufacchi and their colleagues hypothesized that the responses of peripersonal neurons could be associated with assessments of one’s immediate surroundings, specifically in terms of the extent to which different actions would lead to rewards or punishments. To test this hypothesis, they trained ANNs to intercept or avoid objects, then tried to determine whether this resulted in similar body-part-centered responses as those previously observed in the human brain.

“We then proposed a theoretical construct, an ‘egocentric value map,’ which is constructed from groups of peripersonal neurons, forming a more abstract, predictive model of the world near the body that allows rapid adaptation to novel situations,” said Bufacchi. “This idea helped us unify our findings with broader theories in computational neuroscience by framing body-centered responses as part of a flexible, predictive model of the nearby environment.”

After they had created their “egocentric value map,” the researchers compared it to the observations gathered during neuroscience studies performed by multiple labs. The data they compared it to included recordings of the activity of neurons in the brains of macaques, as well as human functional magnetic resonance imaging (fMRI) scans, electroencephalography (EEG) scans and behavioral patterns observed during experiments.

“In brief, we found that the neurons in our artificial agents naturally developed body-part-centered receptive fields that matched empirical findings from biological peripersonal neurons, supporting our theoretical assumptions,” explained Iannetti.

“Specifically, these neurons’ receptive fields expanded with faster-moving stimuli, tool use, and higher-value objects. The networks of artificial neurons also separated into sub-networks specialized for avoidance and interception, mirroring the modularity of both the macaque brain and the egocentric value map that we propose.”

The researchers were ultimately able to demonstrate that a set of peripersonal neurons can in fact create an egocentric map of a primate’s surroundings. They then compared the theoretical framework they had developed to previous interpretations of peripersonal neurons and their function.

“Our theory was the only one to successfully fit extensive experimental data, outperforming alternative explanations and providing a generalizable framework for understanding peripersonal responses,” said Iannetti.

The recent work by Iannetti, Bufacchi and their colleagues contributes to the understanding of peripersonal neurons in the primate brain and how they map out the environment immediately surrounding the body of primates or humans. Yet the insight gathered by the team could soon also help to advance embodied AI agents, robotic systems and prosthetics,

“These findings have potential applications in fields such as neuroprosthetics and human–robot interactions,” explained Iannetti. “For example, robots could simulate egocentric value maps to develop adaptive, context-specific representations of appropriate human interaction distances, making human–robot collaboration more natural and effective.”

The researchers are now planning to build on their findings and continue testing the validity of the framework they introduced. In their next studies, they will test the predictions generated by their computational model and try to address some of its shortcomings.

“For example, the model is currently framed in a reinforcement learning perspective, which lacks explicit parameters for sensory uncertainty,” added Bufacchi. “We will solve this by using different mathematical framings such as active inference, which explicitly incorporates sensory uncertainty and cognitive modeling of the environment. We also plan to collaborate across labs to model richer, more fine-grained and contemporary neuronal data.”

Written for you by our author Ingrid Fadelli, edited by Lisa Lock, and fact-checked and reviewed by Robert Egan—this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive. If this reporting matters to you, please consider a donation (especially monthly). You’ll get an ad-free account as a thank-you.

© 2025 Science X Network

The brains of humans and other primates are known to execute various sophisticated functions, one of which is the representation of the space immediately surrounding the body. This area, also sometimes referred to as “peripersonal space,” is where most interactions between people and their surrounding environment typically take place.

Researchers at Chinese Academy of Sciences, Italian Institute of Technology (IIT) and other institutes recently investigated the neural processes through which the brain represents the area around the body, using brain-inspired computational models. Their findings, published in Nature Neuroscience, suggest that receptive fields surrounding different parts of the body contribute to building a modular model of the space immediately surrounding a person or artificial intelligence (AI) agent.

“Our journey into this field began truly serendipitously, during unfunded experiments done purely out of curiosity,” Giandomenico Iannetti, senior author of the paper, told Medical Xpress. “We discovered that the hand-blink reflex, which is evoked by electrically shocking the hand, was strongly modulated by the position of the hand with respect to the eye.

“We soon realized that this blink reflex behaved much like so-called peripersonal neurons, which are neurons that respond to objects near the body. As we got more familiar with the literature on this type of neuron, however, we noticed that the existing theoretical explanations of their activity fail to explain quite a lot of their properties, such as their modulation by stimulus valence, speed, and motor repertoire.”

Rather than collecting new data, which could then be added to the extensive and disjointed data collected during previous studies, Iannetti and his colleagues set out to develop a new quantitative framework that clarifies why the peripersonal neurons observed in earlier experiments exist and how they work. This framework could then be integrated with existing neuroscience theories.

To develop their framework, they employed artificial neural networks (ANNs) trained via reinforcement learning. These are brain-inspired computational models that can learn to complete various tasks with good accuracy, emulating the connections between neurons.

“In simple terms, we built computer simulations of simplified ‘animals,’ which learned through trial and error to choose actions based on how much reward or punishment those actions would bring over time,” explained Rory John Bufacchi, first author of the paper.

“Our approach involved three main steps. First, our key insight was that peripersonal responses might simply reflect the value of potential actions: whether reaching out to or dodging environmental objects would lead to rewards or punishments.”

Iannetti, Bufacchi and their colleagues hypothesized that the responses of peripersonal neurons could be associated with assessments of one’s immediate surroundings, specifically in terms of the extent to which different actions would lead to rewards or punishments. To test this hypothesis, they trained ANNs to intercept or avoid objects, then tried to determine whether this resulted in similar body-part-centered responses as those previously observed in the human brain.

“We then proposed a theoretical construct, an ‘egocentric value map,’ which is constructed from groups of peripersonal neurons, forming a more abstract, predictive model of the world near the body that allows rapid adaptation to novel situations,” said Bufacchi. “This idea helped us unify our findings with broader theories in computational neuroscience by framing body-centered responses as part of a flexible, predictive model of the nearby environment.”

After they had created their “egocentric value map,” the researchers compared it to the observations gathered during neuroscience studies performed by multiple labs. The data they compared it to included recordings of the activity of neurons in the brains of macaques, as well as human functional magnetic resonance imaging (fMRI) scans, electroencephalography (EEG) scans and behavioral patterns observed during experiments.

“In brief, we found that the neurons in our artificial agents naturally developed body-part-centered receptive fields that matched empirical findings from biological peripersonal neurons, supporting our theoretical assumptions,” explained Iannetti.

“Specifically, these neurons’ receptive fields expanded with faster-moving stimuli, tool use, and higher-value objects. The networks of artificial neurons also separated into sub-networks specialized for avoidance and interception, mirroring the modularity of both the macaque brain and the egocentric value map that we propose.”

The researchers were ultimately able to demonstrate that a set of peripersonal neurons can in fact create an egocentric map of a primate’s surroundings. They then compared the theoretical framework they had developed to previous interpretations of peripersonal neurons and their function.

“Our theory was the only one to successfully fit extensive experimental data, outperforming alternative explanations and providing a generalizable framework for understanding peripersonal responses,” said Iannetti.

The recent work by Iannetti, Bufacchi and their colleagues contributes to the understanding of peripersonal neurons in the primate brain and how they map out the environment immediately surrounding the body of primates or humans. Yet the insight gathered by the team could soon also help to advance embodied AI agents, robotic systems and prosthetics,

“These findings have potential applications in fields such as neuroprosthetics and human–robot interactions,” explained Iannetti. “For example, robots could simulate egocentric value maps to develop adaptive, context-specific representations of appropriate human interaction distances, making human–robot collaboration more natural and effective.”

The researchers are now planning to build on their findings and continue testing the validity of the framework they introduced. In their next studies, they will test the predictions generated by their computational model and try to address some of its shortcomings.

“For example, the model is currently framed in a reinforcement learning perspective, which lacks explicit parameters for sensory uncertainty,” added Bufacchi. “We will solve this by using different mathematical framings such as active inference, which explicitly incorporates sensory uncertainty and cognitive modeling of the environment. We also plan to collaborate across labs to model richer, more fine-grained and contemporary neuronal data.”

Written for you by our author Ingrid Fadelli, edited by Lisa Lock, and fact-checked and reviewed by Robert Egan—this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive. If this reporting matters to you, please consider a donation (especially monthly). You’ll get an ad-free account as a thank-you.

© 2025 Science X Network