The goal of this project is to implement SLAM on the Turtlebot3 Waffle Pi in Gazebo. The two key questions that inspired the development of SLAM are:
- Where is the robot in the world?
- Where are the stationary/moving items in the world?
The SLAM problem has been solved in a probabilistic framework using the Bayes Filter. I focus on implementing the Extended Kalman Filter(EKF) to perform the following:
- Localisation only
- Localization and Mapping with unknown correspondences. I implement the Maximum Likelihood estimator to determine correspondences.
This implementation of the EKF SLAM is still not very robust and is currently being tested and updated.
More technical details on the working of the EKF implementation in the waffle_slam package can be found here
- Ubuntu 16.04
- ROS Kinetic
Open the terminal and run the following commands.
sudo apt-get update
sudo apt-get install ros-kinetic-joy ros-kinetic-teleop-twist-joy ros-kinetic-teleop-twist-keyboard ros-kinetic-laser-proc ros-kinetic-rgbd-launch ros-kinetic-depthimage-to-laserscan ros-kinetic-rosserial-arduino ros-kinetic-rosserial-python ros-kinetic-rosserial-server ros-kinetic-rosserial-client ros-kinetic-rosserial-msgs ros-kinetic-amcl ros-kinetic-map-server ros-kinetic-move-base ros-kinetic-urdf ros-kinetic-xacro ros-kinetic-compressed-image-transport ros-kinetic-rqt-image-view ros-kinetic-gmapping ros-kinetic-navigation ros-kinetic-interactive-markers
The eventual aim of this project is to run the EKF package on a Turtlebot3 in the real world.
After installing the ROS dependent packages, you can simply clone this repository into your workspace and run the catkin_make
command.
cd
into your workspace and source your setup file: source devel/setup.bash
-
To run the path tracking EKF localization implementation, run
roslaunch waffle_slam localization.launch
-
To run the EKF SLAM implementation run the command
roslaunch waffle_slam slam.launch
. This open up both Rviz and Gazebo. On your terminal, you can use the keyboard keys w, a, s and d to navigate while the robot builds an estimate of it's pose and of the detected features in the environment
I am currently working on improving the EKF SLAM so that it is more robust and can be used on a robot in the real world. In addition, I will implement the popular Particle Filter of Localization and Mapping.
- SLAM is a key step for proper Navigation and Planning in mobile autonomous systems
- Tracking: Initial pose is known and robot has to track it's position and orientation
- Global Localisation: Unknown initial pose
- Kidnapped Robot: Robot in operation is taken to an arbitrary position and has to localize.
- Sensor Fusion: Combining data and information from different sensor measurements to get a more accurate state estimate
- Data Association: Linking uncertain measurements with known feature, path or track.
- Loop Closure Detection: Correctly asserting that a robot has visited a location previously.
- Registration: Associating collection of data into a known coordinate system
- Kalman Filter, Extended Kalmn Filter and Unscented Kalman filter: Family of gaussian filters
- Histogram filter or Grid Localisation algorithm
- Markov Localisation: Utilizes discrete probability distribution as representation of state space and updates the probabilities with each iteration
- Particle filter or Monte Carlo localisation
- Assumes linear motion and measurement models
- Assumes a unimodal gaussian
- Not suited to localisation problem of a wheeled robot. Extended Kalman filter is more appropriate for wheeled robots.
- Does not asssume linear measurement or motion models
- Linearization of non-linear function is necessary