Summary: Researchers developed CapuchinAI, an open-source, battery-powered platform that automates cognitive studies of wild primates using facial recognition and touchscreen interaction.
Field-tested in Costa Rica’s Taboga Forest Reserve, the system uses YOLO computer vision to identify wild white-faced capuchins with 97 percent accuracy, deliver personalized touchscreen tasks via a Raspberry Pi microcomputer, and dispense food rewards automatically.
By merging lab-level experimental control with real-world ecological contexts, CapuchinAI offers a low-cost, scalable framework for mapping individual cognitive variations across wild animal populations without removing them from their natural habitat.
Key Facts
- Closed-Loop Automated Testing: CapuchinAI integrates facial recognition vision, interactive touchscreen tasks, and an automated motor-driven food dispenser into a unified, battery-operated field box running on a single Raspberry Pi.
- Computer Vision Accuracy: Trained on GoPro video datasets using the open-source YOLO (You Only Look Once) architecture, the vision pipeline identified specific individual capuchins in wild environments with 97 percent accuracy.
- Rapid Habituation and Learning: Wild white-faced capuchins (Cebus capucinus) in Costa Rica quickly habituated to the physical platform, learning touchscreen-reward associations spontaneously without human intervention or hand-testing.
- Tailored Experimental Protocols: The software detects individual identities to deliver personalized cognitive tasks across four main domains: learning speed, impulse control, cognitive flexibility, and working or long-term memory.
- Resource Management and Anti-Monopolization: The automated software tracks individual participation limits per session, preventing dominant group members from monopolizing the testing box and ensuring balanced data collection across the troop.
Source: Emory University
Scientists created an AI system that uses facial recognition and real-time, touchscreen testing to automate cognitive studies of capuchin monkeys in the wild.
The American Journal of Primatology published a proof-of-concept for the novel method — dubbed CapuchinAI — developed by researchers at Emory University and Georgia Institute of Technology.
The article provides a roadmap for the first scalable, systematic way to evaluate and monitor the cognitive abilities of wild primates.
“The primate brain didn’t evolve in a lab, it evolved in complex, competitive environments,” says Marcela Benítez, Emory assistant professor of anthropology and senior author of the paper. “Yet primate cognition is rarely studied in the wild because the experimental control needed to measure cognition is difficult in unpredictable environments.”
“CapuchinAI” integrates a compact, battery-powered computing system into a field-research platform. The system identifies an approaching monkey, presents a learning task tailored to that individual on a touchscreen, and automatically delivers a food reward if the monkey performs the task correctly.
Field tests of the prototype in the Taboga Forest Reserve of Costa Rica found that CapuchinAI identifies individual capuchins with 97% accuracy, following training on still images and videos. Wild capuchins rapidly habituated and learned touchscreen-reward associations, demonstrating that the system provides a scalable field method for cognitive testing, while also mapping individual differences across tasks.
‘The minds behind the personalities’
“This project builds on the legacy of Frans de Waal,” says Federico Sánchez Vargas, first author of the paper and an Emory PhD student of anthropology.
De Waal pioneered studies of animal cognition as director of Emory’s Living Links Center for the Advanced Study of Ape and Human Evolution, while also writing best-selling books that helped popularize the field. He passed away in 2024.
In addition to lab-based behavioral experiments, de Waal “gave us intimate, beautiful portraits of the lives of primates, treating them as individuals,” Sánchez Vargas says.
“Our AI method allows us to more deeply understand individuals that we already have data on through field observation. We can now automate cognitive testing of them and quantify the findings. It’s a way of getting into the minds behind the personalities. Studying individuals in their natural environments, where there are tons of variations in their life experiences, lets us learn how environmental influences shaped them.”
Co-authors of the paper include Jacob Abernethy, Georgia Tech associate professor of computer science; and Sai Rakshith Potluri, a former Georgia Tech graduate research assistant who is now a software engineer at ExtraHop in Seattle.
The open-source paper includes a guide to the computer coding developed for the system, along with a blueprint to build a low-tech, low-cost field-research platform and to integrate all the components into a closed-loop pipeline. The authors hope other scientists will adapt their AI method to generate cognitive data spanning different species of wild primates, living in a range of environments.
Bridging lab and field
Benítez’ work lies at the intersection of anthropology, psychology and evolutionary biology. She studies cooperation and other social behaviors in monkeys, including a captive population of tufted capuchins in a laboratory and wild, white-faced capuchins in the Taboga Forest Reserve of northeastern Costa Rica. She is a co-director of Capuchinos de Taboga, a research project launched in 2017 in collaboration with the Universidad Nacional Técnica of Costa Rica.
Experiments with animals in labs can be tightly controlled. The results, however, may be skewed since the animal is not interacting within its natural environment. Animal behavior experiments in the wild provide valid social and ecological contexts but they are challenging to design and to control.
“I’m trying to bridge that gap,” Benítez says.
She decided to investigate the potential of AI to achieve this aim.
A seed grant from Emory’s AI.Humanities program launched a collaboration between Benítez and Abernethy to design an AI model for facial recognition of wild capuchins.
Abernethy and Potluri used an open-source software known as YOLO (You Only Look Once) to develop a model to run on a laptop. The researchers trained the model on high-quality GoPro imagery of six wild capuchins interacting with testing platforms in Taboga.
Emory and Georgia Tech undergraduates performed the labor-intensive task of digitally placing “bounding boxes” to frame the faces of the monkeys in thousands of still images and videos tagged with their identities.
The result was a facial-recognition system that could identify these six capuchins with 97% accuracy from static images, video and live footage in the field.
DIY ingenuity
Sánchez Vargas, who joined Emory as a graduate student in 2023, took on the next phase of the AI project: figuring out how to integrate the facial-recognition model into a field-friendly, scalable computer interface that could present tasks to interacting capuchins and dispense food rewards.
“The model was great at identifying six monkeys, but there are 100 capuchins at the Costa Rica field site,” he says. “And we didn’t have high quality video of all of these individuals, which is needed to train the model.”
Sánchez Vargas’ undergraduate degrees are in evolutionary biology and psychology. He is not an expert computer coder, but he dove into the challenge, using Python programming language to simplify and change the parameters of the original facial-recognition model.
“I essentially dumbed it down,” he says, so that instead of classifying individual capuchin faces, the system recognized any capuchin monkey — and only capuchins.
The idea, he explains, was to enable the system to trigger a webcam to record video whenever a capuchin approached a computer touchscreen, rapidly generating a larger, more up-to-date dataset of faces from the interacting monkeys. The resulting videos could then be used to keep training the facial-recognition model, expanding its face-recognition repertoire.
A second Python script Sánchez Vargas developed uses a program called “pygame” to facilitate interactive stimuli for use in games or cognitive testing. The researchers created a simple stimulus to habituate the monkeys to the system: a blue-square covering the computer touchscreen that records when a capuchin touches it. At the monkey’s touch, the script signals a motor circuit to dispense a food reward.
The two Python scripts are integrated to run simultaneously on a Raspberry Pi, a tiny computer about half the size of an iPhone. The entire system can run for eight hours on a lightweight battery pack before it needs recharging.
“It was important to us that our technology be both low cost and ecologically friendly,” Sánchez Vargas says.
Build it and they will come
The next challenge was to create a wildlife-proof, weather-proof platform to house all the components of the system.
“We built it in my garage,” Benítez says.
She and Sánchez Vargas bought planks of pine, deck sealant, rubber insulating strips and plastic piping from Home Depot.
“A person helping us asked, ‘What is the project you’re working on?’” Benítez says. “We didn’t go into it; it would have been a bit complicated to explain.”
Emory TechLab helped Sánchez Vargas create a 3D printed, plastic food dispenser.
The two researchers put together a wooden box, only about 20 inches tall, to house the system’s components: a webcam, a computer touchscreen, the Raspberry Pi, the food dispenser and a rotary motor to power the dispenser.
“It was a lot of work but also a lot of fun,” Sánchez Vargas says. “One of the best parts of working in animal cognition is getting creative, trying to put yourself into the mind of the animal you’re studying, so you can develop a good way to engage them in experiments.”
Finally, it was time to test their creation in the field.
In the field with dreams
Upon arriving at the Capuchinos de Taboga research facility in Costa Rica, Sánchez Vargas remembers feeling a few last-minute jitters. Would the prototype function as envisioned?
“We had to get up super early, at four in the morning,” he says, “and drive a long way down this bumpy road to get to the research site.”
The team secured the CapuchinAI box to a platform, loaded its food dispenser with dried slices of forest banana, and waited nearby.
“For the first couple of days that we did this, no capuchins visited,” Sánchez Vargas says. “I was beginning to worry.”
The third day, however, a large male capuchin named Trompudo, which means “big snout,” could not resist the scent of banana.
“He climbs on the box and starts slapping the back of it,” Sánchez Vargas recalls. “Finally, he slaps the touchscreen and a banana slice pops out.”
Trompudo gobbled up the food then put his hand on the screen again. Another banana slice popped out.
“It was almost like you could see him realizing, ‘Ah, that’s what you have to do, touch the screen!’” Sánchez Vargas says. “It was amazing to watch an individual learn something so quickly.”
Different learning styles
After Tompudo broke the ice, more monkeys began engaging with CapuchinAI.
“Even I was surprised by their enthusiasm,” Sánchez Vargas says.
The researchers are already noting individual differences within the 16 monkeys who engaged with the CapuchinAI prototype during the pilot phase.
Some capuchins are remarkably fast learners. Others take more time to figure out that they need to touch the screen to get a food reward. Individuals who investigated the box with their lips learned to kiss the screen to get a banana slice. Then there are the late adopters. They hang back and watch their friends interact with the box, seeing how the apparatus works before approaching it.
The box stood up to the occasional aggressive moves of capuchins trying to bust it open.
The facial-recognition software prevents the system from activating when other wildlife approach, but some animals still try to tinker with the box, including coatis. Members of the raccoon family, coatis are notorious for their ability to break into manmade containers and even houses. The box passed the coati test.
The research team is now updating its facial-recognition model, training it on the recorded videos of the 16 capuchins from the pilot phase.
They are developing cognitive experiments for four broad domains of cognition, including tests of individual capuchins’ ability to learn, their level of impulse control, their cognitive flexibility, and their skill at both short- and long-term memory.
If the model recognizes a capuchin it was trained on, it will present a specific cognitive test on the touchscreen, depending on what “level” an individual is at in the testing sequence. If the individual is unfamiliar, the model will assign the baseline habituation stimulus — touch the screen to get a reward.
Machine vision allows the model to quickly shift tests from one identified capuchin to another, so multiple individuals can participate. The system is also programmed to limit the number of food rewards an individual can receive during a session, discouraging a dominant capuchin from monopolizing the platform.
A powerful new tool
The researchers can draw from decades of accumulated observational data on the life histories of individuals in the capuchin population of Taboga Forest Reserve. CapuchinAI allows them to learn how variations in the capuchins’ lives may have influenced their cognition.
“We can explore outstanding questions about how the environment, individual experiences and behaviors connect to cognitive abilities,” Benítez says. “Why are some individuals better at some tasks than others? How do different individuals adapt to different situations? How do different cognitive strategies relate to fitness?”
“Frans de Waal said that you cannot study cognition if you don’t understand the animals,” adds Sánchez Vargas. “Our AI methodology doesn’t replace the need for human researchers in the field. It’s essential to have rich, observational datasets gathered by people working on the ground.”
The unusually large brains of primates and their advanced cognition compared to other animals make them key models for the study of how brains and minds evolve and adapt. The researchers believe their AI model is adaptable to other species of wild primates living in the wild for which scientists have recorded individual life histories.
“It’s a powerful new tool in the primatology toolkit,” Benítez says.
Funding: In addition to seed funding from AI.Humanities, the project was supported by the National Institute on Drug Abuse of the U.S. National Institutes of Health (R34DA061925), the U.S. National Science Foundation (BCS-2127373), the Lewis and Clark Fund for Exploration and Field Research; and the Emory Center for Mind, Brain and Culture.
Key Questions Answered:
A: Primate brains evolved to navigate highly complex, unpredictable social and ecological landscapes rather than controlled laboratory cages. While lab studies offer tight experimental variable control, they remove animals from natural social dynamics and varied life histories. CapuchinAI brings precise lab-style experimental control directly into the wild, recording how natural environments shape individual cognitive strategies.
A: Because the system identifies individual animals in real time, the Python control software tracks how many rewards each monkey receives during a given session. Once an individual reaches its designated reward threshold, the touchscreen temporarily halts task presentation for that specific monkey, encouraging other lower-ranking or less assertive group members to interact with the device.
A: The platform is designed using accessible, low-cost components. The computing core uses a lightweight battery pack powering a Raspberry Pi microcomputer connected to a standard webcam, a low-cost touchscreen, and a 3D-printed plastic rotary food dispenser housed inside a weatherized, monkey-proof pine enclosure.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this neurodevelopment and mental health research news
Author: Carol Clark
Source: Emory University
Contact: Carol Clark – Emory University
Image: The image is credited to the researchers
Original Research: Open access.
“CapuchinAI 1.0: Development of a Machine Learning-Based Touchscreen Paradigm to Test Cognition in Wild Capuchins” by Federico Sánchez Vargas, Sai Rakshith Potluri, Jacob Abernethy, Marcela E. Benítez. American Journal of Primatology
DOI:10.1002/ajp.70194
Abstract
CapuchinAI 1.0: Development of a Machine Learning-Based Touchscreen Paradigm to Test Cognition in Wild Capuchins
Advancing the study of primate cognition requires methods that preserve ecological validity while enabling the experimental control typical of laboratory research.
We introduce CapuchinAI v1.0., a field-deployable touchscreen system currently integrating real-time species recognition with automated cognitive testing, providing a novel methodology for studying cognition in wild primates. Our approach combines an adapted version of a high-performing YOLOv7-based facial recognition model (MultipleCapuchins) with a portable Raspberry Pi touchscreen–reward apparatus designed for automated operation in natural habitats.
The system detects approaching capuchins, initiates video recording, presents stimuli (in our initial deployment, a blue screen that rewards all touches), and dispenses food rewards. During a 2-week presentation to two habituated groups of wild white-faced capuchins (Cebus imitator) at the Taboga Forest Reserve, 16 individuals voluntarily interacted with the apparatus, 10 triggered rewards, and 8 formed and retained robust screen–reward associations.
The rapid habituation and learning rates demonstrate the feasibility of deploying AI-mediated cognitive experiments in the wild. Ongoing development of CapuchinAI aims to address several long-standing challenges in field cognition research, with the goal of enabling: (1) autonomous, individualized task administration without researcher intervention; (2) standardized, repeatable trials across individuals and sessions; (3) scalable deployment across groups and sites; and (4) parallel data collection on behavior, identity, and performance.
This methodology provides a blueprint for integrating machine learning and touchscreen testing to study within- and between-individual cognitive variation under natural conditions. CapuchinAI represents a significant step toward long-term comparative research on primate cognition, bridging the gap between lab and field.
