Problem
Build an end-to-end robotic workflow that can classify and pick waste objects reliably in a fixed workspace.
Project
GreenArm: Vision Guided Robotic Waste Sorting
ROS2 perception-to-manipulation pipeline using YOLOv8, OpenCV transforms, and ArUco calibration for precise robotic sorting.
Approach
I helped build a ROS2 architecture connecting detection, calibration, coordinate mapping, and arm control.
Result
The system achieved repeatable placement precision once calibration and transform caching were handled carefully.
Summary
GreenArm was an integration-heavy robotics project rather than a single-model exercise.
What mattered
- Stable calibration with ArUco markers.
- Reliable pixel-to-robot coordinate conversion.
- Better handling of detection-to-actuation latency.
- Clear separation between perception logic and arm control.