View Code
2025.07 - 2025.08 Team Project ROS 2 / Embedded AI

Roomie: Hotel Delivery Robot

An autonomous hotel service robot capable of inter-floor navigation using a robot arm to operate elevators. Delivers room service and provides guided wayfinding for guests.

Repository Back to Projects

Project Overview

Roomie is designed to automate repetitive hotel operations, reducing staff workload while providing a novel experience for guests. Its standout feature is the ability to independently navigate between floors by physically pressing elevator buttons with an attached 4-DOF robot arm.

Key Engineering Contributions

Project Leadership & Architecture

Led the development team and designed the entire ROS 2 package structure. Architected the central Finite State Machine (FSM) that manages complex asynchronous tasks like elevator navigation, room docking, and service interactions.

System Architecture

Vision AI Pipeline

Engineered a multi-stage vision pipeline for reliable elevator operation. Cascaded YOLOv8 for panel detection, a custom CNN for button classification, and EasyOCR for floor verification, achieving 95% success rate in real-world trials.

Vision Processor

Interactive Robot GUI

Developed the robot's onboard display interface using PyQt and ROS 2. The GUI provides real-time state visualization (e.g., "Going to 5th Floor") and intuitive touch interactions for guests and staff.

Robot Interface

Core Technology

  • Elevator manipulation: Gaussian velocity profiling for smooth arm control; Coordinate mapping from camera to button space.
  • Navigation: Nav2-based path planning with dynamic obstacle avoidance using depth cameras.
  • Vision Pipeline: YOLOv8n for object/obstacle detection -> CNN for button classification -> EasyOCR for floor number reading.
  • Embedded Control: micro-ROS on ESP32 for managing ultrasonic sensors, RFID readers, and LED indicators.

Tech Stack

ROS 2 (Humble) Nav2 YOLOv8 DeepSORT PyQt6 ESP32 (micro-ROS) Python & C++

Technical Challenges & Solutions

Arm Vibration & Accuracy

Problem: Rapid arm movements caused vibrations, leading to failed button presses.
Solution: Applied Gaussian velocity and acceleration profiles to smooth out trajectories, significantly improving stability.

Navigation in Narrow Corridors

Problem: Standard path planners struggled with dynamic obstacles in narrow hotel hallways.
Solution: Implemented waypoint-based local detours and optimized the A* global planner parameters.

Button Recognition Reliability

Problem: False positives in button detection.
Solution: Created a multi-stage pipeline: YOLO detects the panel, CNN classifies the button type, and EasyOCR verifies the floor number.