turtlebot-detect-slam

This repository contains documentation and resources for doing visual SLAM (simultaneous localization and mapping) on a TurtleBot3 Waffle Pi and ROS. This repository is a work in progress and will be constantly changing as I research and experiment with different packages and tools.

Setup

Turtlebot Setup

• Follow the setup guide in the ROBOTIS e-manual.

  • Everything in this repository is tested using Ubuntu 20.04 with ROS Noetic on the Remote PC, and the provided Raspbian disk image in the manual (Section 6.2.1.2).
  • The Remote PC you use should have a somewhat powerful Nvidia GPU if you want the segmentation/detection packages to run at a respectable framerate.

Object Detection Setup - YOLOv3

• Follow installation instructions for the darknet ROS package.
  • I found this script helpful for installing OpenCV.
  • CUDA installation guide. Make sure you follow the mandatory post-installation steps!

Optional Setup

• I recommend installing ROS web video server so you can easily look at different image topics in your web browser. Otherwise you can use rqt_image_view or something.

sudo apt-get install ros-melodic-web-video-server

Usage

• Start up ROS master. SSH into the turtlebot to start up its sensors and its camera with the following commands.

roslaunch turtlebot3_bringup turtlebot3_robot.launch

roslaunch raspicam_node camerav2_1280x960.launch

• Run the launch file (on the Remote PC).

roslaunch visual_slam classify_and_locate.launch

• To use custom weights, either edit yolo_v3.launch in the darknet_ros package or make a new launch file that points to the proper weights/configuration files. Make sure to edit the launch file in this package to point to the new darknet_ros launch files as well.

Simulation

This package also has support for Gazebo simulation, with the Turtlebot3 Simulations package as a prerequisite. The launch file opens an empty world by default.

roslaunch visual_slam gazebo_classify_and_locate.launch

Classes in weights files

bevo.weights

• bevo

turtlebot.weights

• turtlebot_burger

multiclass.weights

• box
• bevo
• bottle
• footwear
• turtlebot

multiclass-v3.weights

Updated version of multiclass.weights (no footwear)

• box
• bevo
• bottle
• turtlebot