Home

Perception Pipeline — Wiki

Welcome to the Robots4Sustainability Perception wiki. This repository provides a complete ROS 2 vision pipeline for robotic object detection, pose estimation, and scene understanding — built on top of YOLOv8 and Open3D, designed for the Kinova robotic arm platform.

---

📋 Pages

Page	Description
Architecture Overview	System design, nodes, and the action-server pattern
Setup & Installation	Prerequisites, environment setup, and building
Camera Configuration	RealSense and Kinova camera launch instructions
Vision Pipeline Tasks	All 6 tasks: what they do, how to trigger them
Action Server & Brain Client	How the dispatcher and client work
Lifecycle Node System	Lazy vs eager loading, VRAM management
RViz Visualization	Pre-built configs and topic guide
ROS 2 Topics Reference	Full topic listing with types
Fine-Tuning Models	Dataset, training, and evaluation workflow

---

🤖 Quick Start

# Terminal 1 — Launch all nodes
ros2 launch perception action_launch.py lazy_loading:=true
 
# Terminal 2 — Trigger a task
ros2 run perception brain_client <task_name>

Available tasks: table_height · detect_screws · car_objects · subdoor_pose · detect_screwdriver · place_object

---

🧰 Tech Stack

Component	Technology
Framework	ROS 2 Jazzy
Language	Python 3.12
Object Detection	YOLOv8 (Ultralytics)
OBB Detection	YOLOv8-OBB
3D Processing	Open3D
Camera	Intel RealSense / Kinova Vision
Node Pattern	ROS 2 Lifecycle Nodes