CS3340 - Robotics and Automation | Computer Science and Engineering, University of Moratuwa
An autonomous mobile robot that detects, collects, and places cubes using YOLO-based vision and RGB-D sensing on the Kobuki QBot platform.
In this final project we are building robot(Kobuki-Qbot) that operates in a controlled indoor environment. Starting from a charging station(middle of the arena), the robot searches for cubes, collects them using a maipulator arm, deposits them in a designated drop-off zone, and returns to dock, all without human intervention.
The project is motivated by real-world applications such as:
- Automated sorting in logistics warehouses
- Debris clearance in hazardous environments
- Educational robotics demonstrations
The system is built on three core robotics pillars: Perception, Planning, and Control, integrated via the ROS 2 framework.
| Component | Details |
|---|---|
| Mobile Base | Kobuki QBot (differential-drive) |
| Vision System | Xbox Kinect RGB-D Camera |
| Compute | Raspberry Pi 5 (2GB) |
| Manipulator | Simple gripper arm attached to base |
| Power | Kobuki battery with charging dock + voltage monitoring sensor |
| Layer | Technology |
|---|---|
| Framework | ROS 2 Jazzy |
| Object Detection | YOLOv8 Nano |
| Depth Processing | Point Cloud Library (PCL) |
| Task Sequencing | SMACH State Machine |
| Simulation | Gazebo |
| Localisation Markers | AprilTags |
- Ubuntu 22.04 (or compatible)
- ROS 2 Jazzy installed
- Python 3.10+
- Gazebo (for simulation)
- Use this link to find the yolotraing results and model.
Note that the rasberrypi integration in the following branch.
# Clone the repository
git clone https://github.com/IntellisenseLab/final-project-botzilla
cd botzilla
# Install Python dependencies
pip install -r requirements.txt
# Build the ROS 2 workspace
colcon build --symlink-install
source install/setup.bash# Launch the Gazebo simulation environment
ros2 launch botzilla simulation.launch.py
# In a new terminal, start the main autonomy stack
ros2 launch botzilla botzilla_autonomy.launch.py# Ensure Kobuki and Kinect are connected, then:
ros2 launch botzilla hardware.launch.py
# Start the autonomy stack
ros2 launch botzilla botzilla_autonomy.launch.pyfinal-project-botzilla/
├── botzilla_Workspace/src/
│ ├── botzilla_bringup/
│ ├── botzilla_control/ # drive base control
│ ├── botzilla_perception/ # yolo and apriltag detection
│
├── datasets/ # Cube Dataset
├── docs/ # documents
├── runs/ # Best Fited ML models
├── tests/ # Unit and integration tests
├── Worlds/
├── requirements.txt
└── README.md
- ≥ 80% accuracy in cube detection and localization under varying lighting conditions
- Successful collection and placement of all cubes in the drop-off zone (0.5m × 0.5m)
- Safe return to the charging dock after task completion
- No fault tolerance issues with the gripper arm during manipulation
| Name | Index | Responsibilities |
|---|---|---|
| Mudaliarachchi N.S | 230415H | RGB-D depth sensing, dataset collection, model training |
| H.H. Malavipathirana | 230389E | QBot command configuration, key-command programming, object detection & navigation |
| K.N.B. Abeysundara | 230010L | Raspberry Pi setup, YOLOv8 pipeline, ROS 2 integration, final report |
- Blob Tracking using Kobuki QBot2
- Raspberry Pi 5— Setup & Getting Started
- YOLO on Raspberry Pi
- Kobuki QBot Documentation
- Cube Dataset — Roboflow
The Arena is 300cm*300cm and the charging doc, where robot starts action, is in the middle of the arena. The robot splits arena in to 4 quadrents and those are represented using dashed lines.
Department of Computer Science and Engineering
University of Moratuwa
CS3340 — Robotics and Automation | March 2026