AMR Grasping – Autonomous Object Detection and Manipulation Using the LIMO CoBot

This repository contains the full ROS-based implementation developed for the Master’s thesis
“Autonomous Object Detection and Manipulation Using a Mobile Cobot”
by Tim Yago Nordhoff (TH Köln, 2025).

Overview

The project implements a fully onboard autonomous pipeline that enables a mobile cobot to:

• explore unknown indoor environments
• detect a text-specified target object (open-vocabulary)
• estimate its 3D position
• navigate toward and grasp the object autonomously

All components are designed for real-time execution on embedded hardware.

Platform

• Robot: LIMO COBOT (mobile base with robotic arm)
• Sensors: 2D LiDAR, RGB-D camera, IMU
• Framework: ROS (Noetic), fully onboard execution

Repository Structure

amr-grasping/
├── limo_explore/        # ROS package with exploration, coverage, and object finding nodes
│   ├── scripts/         # Core Python nodes
│   ├── launch/          # Launch files for the system
│   ├── config/          # YAML configuration (costmaps, localization, etc.)
│   ├── rviz/            # Saved RViz configurations
│   ├── CMakeLists.txt
│   └── package.xml
├── traj_imgs/           # Images with the trajectories
├── traj_logs/           # Saved trajectory logs (*.txt)
└── README.md

Launch Files (limo_explore/launch/)

• frontier_demo.launch
  Starts autonomous exploration using the FrontierPlanner, including navigation and camera coverage mapping.

• straight_demo.launch
  Starts autonomous exploration using the StraightPlanner.

• rviz.launch
  Launches RViz with the preconfigured visualization settings from the rviz/ directory.

• object_detection.launch
  Starts only the object detection pipeline (ObjectFinder), without exploration or navigation.

• nav.launch
  Only starts localization, mapping, and move_base.

Key Components

• SLAM & Navigation
  EKF-based odometry fusion, slam_gmapping, move_base

• Exploration
  Camera-based coverage mapping (CamCoverage) and exploration planners

• Perception
  Open-vocabulary object detection using NanoOWL

• Manipulation
  Deterministic grasp strategy (approach → align → close-in → grasp → lift)