REMOP

Built REMOP as a live perception and agent-control framework to generate high-quality training data for humanoid robots operating in household environments. The system combines low-latency websocket streaming, object detection with monocular depth estimation, and an agent loop that turns grounded scene understanding into step-by-step human-executable actions. This architecture enables fast closed-loop data capture for manipulation and tidying tasks while keeping the full pipeline modular across frontend and backend services.