guilhermelawless / turtlebot3_datasets

Datasets captured with the Turtlebot 3 Waffle Pi and MOCAP for ground-truth data.

turtlebot3_datasets

This package provides helper scripts to download and use datasets for the Introduction to Robotics class.

The datasets were captured on a Turtlebot 3 Waffle Pi.

This guide assumes that you have followed the previous tutorials.

Dataset information

A map and rosbag are provided, along with some helper scripts.

The initial bag is still provided, but had a synchronization issue. This has been fixed using fix_stamps.py.

The fixed bag includes:

rosbag info fixed_slam_easy.bag

topics:      /imu                              14805 msgs    : sensor_msgs/Imu            
             /odom                              3252 msgs    : nav_msgs/Odometry          
             /raspicam_node/camera_info         1814 msgs    : sensor_msgs/CameraInfo     
             /raspicam_node/image/compressed    1811 msgs    : sensor_msgs/CompressedImage
             /scan                               626 msgs    : sensor_msgs/LaserScan      
             /tf                               21329 msgs    : tf/tfMessage               
             /tf_static                            1 msg     : tf/tfMessage

The setup includes ground-truth data. This was obtained from a motion capture system at 60Hz. 5 markers were placed on the top layer of the robot, such that the center of the tracked object matched the laser scanner of the robot.

The ground-truth data is provided in the /tf topic, as a transform mocap -> mocap_laser_link. To conform to REP 105 in ROS, the initial transform was obtained at the robot's base footprint frame. The homogeneous transformation matrix at the beginning of the dataset is given by:

The initial transform can be used to connect mocap to odom, map or other fixed frames. This can be done by executing publish_initial_tf.sh.

Images docs/unconnected_tree.svg and docs/connected_tree.svg show the setup in terms of frames, and what happens when we add the mocap -> odom transform. These images have been obtained with the following bash command: rosrun tf2_tools view_frames.py.

The map was obtained using turtlebot3_slam gmapping using the default parameters.

Notices

First, some things to know:

• Use simulation time when reading data from rosbags. If not, then things like rospy.Time.now() will not output the time at which the dataset was recorded. To use simulation time: rosparam set use_sim_time true after roscore. If you were running rviz already, it needs to be restarted. Otherwise, TFs might not be picked up.

• Always play back the rosbag with the --clock option (related to simulation time). --pause and --r RATE can also help.

Steps

1. git clone https://github.com/guilhermelawless/turtlebot3_datasets.git into your ROS workspace

2. Build with catkin: cd $ROS_WORKSPACE/../src && catkin_make && source ~/.bashrc

3. Download the dataset (the map is already in the data directory, this downloads the rosbag): roscd turtlebot3_datasets/scripts && bash download_dataset.sh

4. Run a static transform publisher to connect the ground-truth and robot frames (you can also add as a node to your launch file): rosrun turtlebot3_datasets publish_initial_tf.sh odom # other frames can be used

5. Launch the description launch file: roslaunch turtlebot3_datasets turtlebot3_description.launch

6. Launch map server and/or other algorithms...

7. Play the bag.