MiniSoul Exhibits Evolving AI-Driven Personality on Pocket-Sized Robot

Hackster reports that MiniSoul is a pocket-sized desktop companion robot built by Sritabh Priyadarshi. The hardware centres on an ESP32-S3 SuperMini development board paired with a 0.96-inch OLED display, with an enclosure small enough for keychain use. The firmware implements a custom behavior engine tracking six personality traits: joy, curiosity, fear, anger, sadness, and desire. A capacitive touch surface senses interaction and a small on-device kNN model classifies gentle caresses, playful taps, and more aggressive presses; classified events feed the behavior engine so the robot's emotional state evolves over repeated interactions.

The touch-sensing pipeline consists of capacitive input, lightweight feature extraction, and an on-device kNN classifier that maps touch patterns to discrete interaction types. The behavior engine consumes those interaction labels and adjusts internal trait values over time, resulting in different animations and responses displayed on the OLED. The ESP32-S3 SuperMini provides the compute and small footprint needed for keychain form factor without cloud dependency.

Hobbyist and consumer desktop companions increasingly combine small microcontrollers with on-device machine learning for responsive interaction without cloud dependency. Using an ESP32-class MCU and a compact kNN classifier is a common pattern where latency, privacy, and power consumption favour tiny models running locally. Stateful behavior engines that accumulate interaction history are a low-cost way to produce emergent, personalized UX without large language models or heavy compute.

For practitioners, MiniSoul illustrates how modest hardware and simple ML primitives can create emergent personality without cloud compute, making the project a useful reference for makers building compact robotic companions with configurable behavior. The single Hackster source covers hardware and firmware details; the project does not appear to have a public code repository or expanded documentation at time of publication.