Development and Optimization of the KAI-979 Humanoid Guide Robot: An Offline Intelligent Framework for Sustainable Tourism Eco-Systems

Abstract

This study addresses the connectivity dependence in service robotics by evaluating the V-KAI979 prototype, a bilingual tourist guide named after the 979-meter altitude of Angel Falls. Derived from a full-scale humanoid chassis, this compact, transportable tabletop version utilizes a modular high-density Polylactic Acid (PLA) body. The distributed architecture delegates articular joint management to an external PCA9685 driver via an I2C bus, freeing the core dual-core ESP32 microcontroller’s clock cycles. In Offline Mode, an ultrasonic sensor triggers local linguistic processing, where an optimized TinyML keyword-spotting model trained via Edge Impulse achieves a 91.50% bilingual greeting classification accuracy with a minimal latency of 145 ms. In Online Mode, a custom Python-based API and a dedicated Telegram Bot alleviate computational bottlenecks, expanding the robot's capabilities to process complex historical narratives, regional folklore, indigenous Pemón myths, and traditional music while executing real-time motor dance choreographies. Inferential analysis using the Mann-Whitney U test confirmed a statistically significant latency increase in Online Mode (1240 ms) due to network fluctuations, demonstrating that local edge intelligence guarantees essential operational resilience during wireless infrastructure failures. The mechanical layout remains pre-dimensioned for strategic onboard upgrades, specifically migrating the cloud runtime onto a Raspberry Pi 5 unit coupled with a smart camera for advanced vision