Semantic Memory for Embodied Agent

An embodied AI agent that builds a semantic map of its surroundings and remembers what it has seen. It uses a knowledge graph for landmarks and their relationships, GPT-4 Vision to understand what the camera sees, and LLaMA 70B for reasoning and conversation. Observations get stored in a vector store so the robot can recall past experiences, and it connects to the NAV2 stack to move around in Gazebo.