The CapuchinAI initiative—detailed in The American Journal of Primatology—was launched by researchers from Emory University and the Georgia Institute of Technology to solve a longstanding problem in animal research. Marcela Benítez, an Emory assistant professor, emphasized that primate brains developed in intricate, competitive natural settings rather than laboratory cages. She noted that cognitive studies in the wild are uncommon because the unpredictable natural environment makes establishing strict experimental controls challenging. By moving the technology into the Costa Rican forest, the team established a roadmap for the first systematic, scalable method to monitor the cognitive abilities of wild primates.
System Design and Collaborative Development
The technology relies on a compact, battery-powered computing setup built around a Raspberry Pi microcomputer that can run for up to eight hours on a single charge. The facial recognition model was co-developed by Jacob Abernethy and Sai Rakshith Potluri using the open-source YOLO (You Only Look Once) software.
To train the system, undergraduates from both Emory and Georgia Tech meticulously drew bounding boxes around monkey faces in thousands of GoPro images and videos. As a result, the model achieved a 97% accuracy rate in identifying six specific capuchins from still and live footage. Furthermore, the system was designed to automatically record video whenever a monkey approaches, creating an ever-expanding dataset to continuously improve the facial recognition model.
Benítez noted that the physical enclosure was constructed in her own garage. Alongside researcher Federico Sánchez Vargas, she sourced pine planks, deck sealant, rubber weather-stripping, and plastic pipes from a local hardware store. With the help of the Emory TechLab, they integrated a 3D-printed food dispenser into a compact, 20-inch wooden housing that protected the webcam, touchscreen, and rotary motor from the elements.
