hopkira/k9

Implements a modular architecture using Redis as a shared blackboard and MQTT for pub/sub communication between independent Python processes, enabling future migration to ROS2. The system combines local speech recognition (Mozilla DeepSpeech), GPT-3/3.5 conversation, and custom VITS text-to-speech on a separate CUDA server, with vision processing via OAK-D stereoscopic camera for person detection, obstacle following, and collision avoidance. Hardware integration includes servo controllers, motor encoders, 360° LIDAR, and microcontroller coordination via MQTT, allowing graceful fallback to command-based interaction when cloud connectivity is unavailable.