Autonomous Navigation Project

using a Clearpath Jackal UGV and the ROS Navigation Stack

Nate Kaiser -- MS in Robotics Winter Quarter Project -- Northwestern University

Overview

This package, along with its dependencies, provide everything necessary to navigate the Northwestern Robotics Program's Clearpath Jackal UGV autonomously using laser data from a Velodyne VLP-16 LIDAR.

Specifically, this package uses:

Either the default global_planner or NavFN for global path planning
Either gmapping or hector_slam for mapping/SLAM
AMCL for localizing within a given map
A custom local planner is in the works but is not yet fully implemented

Parameters have been tuned specifically for Northwestern's robot/environment combination, so some tweaks may be necessary if this is ported to a different robot/environment combination.

Installation

Prerequisites

To install all the dependencies, fire up a terminal and run:

sudo apt-get update
sudo apt-get install ros-indigo-velodyne ros-indigo-velodyne-description ros-indigo-velodyne-simulator ros-indigo-jackal-desktop ros-indigo-jackal-gazebo ros-indigo-navigation

Alternatively, I hope to clean everything up and have rosdep install name-of-package functioning properly in the near future. Important Note: you may need to install ros-indigo-velodyne from source, per instructions [here][14] and [here][15], although installing from apt-get seems to be fully functioning when I last checked.

This will need to be done on both the host computer and Jackal's on-board computer, but Jackal should already have ros-indigo-jackal-desktop installed, and has no need for ros-indigo-jackal-gazebo, since Gazebo simulations will only be run on the host computer (though accidentally installing it won't hurt anything).

Note: this assumes you already have a full ROS Indigo installation (ros-indigo-desktop-full) running on Ubuntu 14.04. If not, you may have to install other dependencies that come with the full version (PCL, OpenCV, etc). Also, this package should work on a different ROS version or Linux version/distro, but no testing has been done to verify.

This Package

To install this package:

Clone this repo into a workspace and build^†
SSH into Jackal and repeat the process

This requires Jackal to be set up and connected to the internet. For help with this, Clearpath has provided an excellent walkthrough.

^†I highly recommend using catkin_tools, which has a modernized version of the catkin_make command and other useful command line tools for working with ROS/catkin packages

Instructions For Use

Navigate to your workspace and source your setup file: source devel/setup.bash. If you're running on the real robot, skip step 1, as it just launches the robot simulation. Choose only one of steps 4, 5, or 6, depending on what you want to do.

roslaunch nu_jackal_autonav simulate.launch

This loads a URDF of Northwestern's Jackal configuration to the parameter server, starts up a Gazebo session with Clearpath's tutorial map, and spawns Jackal into it. By default, no GUI is shown, but can be shown by appending gui:=true to the end of the roslaunch command. See the beginning of the launch file for other helpful command line arguments.
roslaunch nu_jackal_autonav pointcloud_filter.launch

This command eliminates the z drift between the odom and base_link frames and starts a nodelet manager to handle all of the pointcloud filtering. The outputs are a new tf frame named odom_corrected and a cascaded series of new pointcloud topics, the important one being /velodyne_points/for_costmap. The individual filters are:
- a cropbox to eliminate any points further than 4 meters from Jackal in the x- and y-directions, and any points below 0.0 or above 0.4 meters in z
- a statistical outlier filter (using pcl/RadiusOutlierRemoval)
- a voxel grid downsampler
- a custom filter which eliminates any points too close to the robot or within 1.5° of the ground
roslaunch nu_jackal_autonav laser_scan_operations.launch

I added some functionality and broke this node out as a separate entity instead of including it in multiple launch files. It produces an emulated LaserScan topic, runs an instance of laser_scan_matcher, and converts the output into an odometry topic for use by the the main robot_localization running on Jackal, which should help improve the odometry estimates.
roslaunch nu_jackal_autonav odom_navigation.launch

This node is similar to the file in the jackal_navigation package with the same name. This puts Jackal in pure odometric navigation mode. Under the hood it just runs move_base and loads the correct parameters from yaml files stored in the params directory. All you have to do is send Jackal goal poses using RViz and watch as it navigates there autonomously.
roslaunch nu_jackal_autonav gmapping.launch

Again, this is similar to its jackal_navigation counterpart. It starts gmapping and move_base with the correct parameters. Run this to make a map of whatever environment(s) Jackal will be navigating in. Drive around using the joystick controller until you're satisfied with the costmap (viewable in RViz). Only 1 map is needed for each environment. Also, please note you'll need to save the map once you're satisfied with it by running rosrun map_server map_saver -f map-name-goes-here in a separate terminal.
roslaunch nu_jackal_autonav amcl.launch

This is the heart of Jackal's navigational capabilities. When working correctly, this node estimates the robot's pose in the map. The launch file starts AMCL and move_base. You'll need to modify the map_file parameter to point to the map you created in the last step (stored in maps directory).
roslaunch nu_jackal_autonav view_rviz.launch

This runs one of two RViz sessions. If you're running gmapping_demo or amcl_demo, you'll need to append config:=map to the launch command. With this running, you can:
- send a pose estimate to AMCL
- send a navigation goal to move_base
- control the robot using convenient and intuitive sliders
- view the full Velodyne pointcloud
- view the pointcloud used for determining costmap values
- view the particle array of the particle filter
- view the local and global costmaps
- view the camera feed
- view the local and global paths
- view the emulated laser scan (used by gmapping/AMCL)

Future Work

Currently, I'm working on a custom local planner plugin to replace the default dwa_local_planner, since it's not ideal for this specific robot setup. I plan to release as a separate package when it's complete.

I'm also working on an RViz plugin for canceling navigation goals, but haven't had much time to work on this yet.

Other Considerations

you might need a decent, dedicated router for fast and uninterrupted data transfer
if the laser location is changed, it will need to be updated in the URDF
all parameters were set for vinyl lab floors, which are somewhat slick - different environments may require retuning
nearby obstacles were always close proximity when developing, so parameters may need adjusting if this is used in a more open area
this was developed in a highly dynamic environment, which may also require retuning some parameters as well
most of the modifiable parameters are stored in yaml or launch files in the params or launch directories, respectively
if navigation is acting funny or not working at all, check that base_local_planner isn't set to custom_planner/CustomPlanner in launch/include/move_base.launch, since I'm not finished developing it yet
OpenCV 2.4 was used, I have not checked compatibility with OpenCV 3
PCL 1.7 was used, I have not checked compatibility with newer versions

njkaiser/nu_jackal_autonav