Feature-based Kalman Filter SLAM in ROS

Overview

This project was done as a collection of homework assignments for ME 495 Sensing, Navigation, and Machine Learning For Robotics at Northwestern. The task was to build a several ROS packages to implement control and SLAM (simultaneous localization and mapping) for a turtlebot3 burger robot. The repo is not made available since this is class homework.

Package list

This repository consists of several ROS packages.

• nuturtle_description - contains URDF files, meshes, and launch file for launching and displaying the turtlebot3 burger in rviz.
• rigid2d - a package for odometry of a differential drive robot, with a library for rigid body transformations in SE(2),
• trect - a package for controlling a turtlesim turtle to move along a rectangular path.
• nuturtle_robot - a package for interfacing with a physical turtlebot3 burger robot. Contains launch files for launching remote nodes, and the turtle_interface_node for low-level commands.
• nurtlesim - a package that provides a simulator for the turtlebot and SLAM in rviz
• nuslam - a package that implements Feature-Based Kalman Filter SLAM for the turtlebot.

Description of packages

rigid2d - Rigid 2D transformation Library

A library for handling transformations in SE(2), as well as handling odometry calculations for a differential drive robot. See this document for derivations of equations used in the diff_drive.cpp library implementation file.

Library implementation files

• rigid2d.cpp
  • usage: include rigid2d/rigid2d.hpp
  • 2D rigid body transformations
• diff_drive.cpp
  • usage: include rigid2d/diff_drive.hpp
  • kinematics of a differential drive robot

Nodes

• rigid2d_fake_turtle_node in fake_turtle.cpp
• rigid2d_odometer_node in odometer.cpp

nuturtle_robot - turtlebot3 low-level controller

This package provides functionality to interface with low-level commands on the turtlebot, as well as launch files to start basic nodes on the turtlebot to allow interfacing with ROS.

Circle path test After following a clockwise and counterclockwise circluar path several times and stopping at the initial position, the odometry location of the robot was (0.2667, 0.13376, -3.683e-03).

nurtlesim - simulator for turtlebot3 burger in tube world in rviz

This packages provides a simulator for the turtlebot and slam in rviz. The simulator simulates the robot kinematics and a sensor that detects the relative x, y positions of landmarks and a landmark id. Landmarks are modelled as tubes (approx 10cm diameter cylinders).

nuslam - Feature-based SLAM

This package implements Feature-Based Kalman Filter SLAM for a turtlebot in the nurtlesim simulation. It also provides a library for handling the SLAM algorithm calculations for a differential drive robot.

Library implementation files

• nuslam_lib.cpp
  • usage: include nuslam/nuslam_lib.hpp
  • Feature-based Kalman Filter SLAM calculation helper functions
  • Data association using mahalanobis distance
  • A maximum of 10 landmarks can be stored
• circle_fitting.cpp
  • usage: include nuslam/circle_fitting.hpp
  • Circle fitting with circle regression
  • Circle classification

Nodes

• nuslam_node in slam.cpp - SLAM with known data association
• landmarks_node in landmarks.cpp - circle fitting and classification
• nuslam_unknown_node in slam_unknown.cpp - SLAM with unknown data association

Demonstration of SLAM with known data association

• The robot was driven around in simulation such that all the landmarks were encountered during the run.
• ground truth landmarks are shown in white
• landmarks detected by the sensor are shown in red
• landmarks according to slam (the map) are shown in blue
• trajectory of the actual robot is shown in green
• trajectory of the robot according to SLAM is shown in orange
• trajectory of the robot according to odometry is shown in white
• turtle (actual turtle position frame)

The result for default R and Q is shown:

Top view:

Perspective view:

Demonstration of Slam with unknown data association

• The robot was driven around in simulation such that all the landmarks were encountered during the run.
• robot is driven at max linear speed 0.01m/s and max angular speed 0.1rad/s
• ground truth landmarks are shown in white
• landmarks detected by the sensor are shown in red
• landmarks according to slam (the map) are shown in blue
• trajectory of the actual robot is shown in green
• trajectory of the robot according to SLAM is shown in orange
• trajectory of the robot according to odometry is shown in white
• turtle (actual turtle position frame)

A screencast of the robot driving around in the environment is shown: