mushr_pf.jl: Particle Filter

Install

Note: If you are using a MuSHR image, you can skip steps 1 & 5

1. Install Julia, there are 2 ways to do it:

   • bash -ci "$(curl -fsSL https://raw.githubusercontent.com/colinxs/jill/feat/detect_arch/jill.sh)"
   • To install, follow the directions here. There's nothing to sudo apt-get, just a binary that you download and add to your PATH environment variable.

2. Clone this repo into your ROS workspace: cd ~/catkin_ws/src && git clone https://github.com/prl-mushr/mushr_pf.git

3. The library used to interoperate with Python (PyCall) needs to be linked to the same Python binary that you use with ROS. By default this is the result of executing which python on the command line. This can be an issue if you perform the next step with an Anaconda environment activated. To avoid this run:
   export PYTHON=$(which python) > ~/.bashrc and re-source your workspace . ~/.bashrc
   If you ever want to switch python versions, just change the PYTHON variable & re-run julia -e 'using Pkg; Pkg.build("PyCall").

4. From the root directory of the repo run cd ./deps && ./jetson_install.sh. The script will prompt you for sudo permissions, but do not run the script itself with sudo. You may see an error about the build for Arpack failing; don't worry, that's what the patch is for :). This builds all necessary dependencies listed in Project.toml & Manifest.toml.

5. If not installed already, install the sensor_msgs, geometry_msgs, nav_msgs, vesc_msgs, and ackermann_msgs packages. The MuSHR images all have these installed.

6. cd ~/catkin_ws/ && catkin make
   (although like Python, there is nothing to build, this step just sets up your environment variables) and run source ~/catkin_ws/devel/setup.bash

7. Run roslaunch mushr_pf ParticleFilter.launch

8. The node will compile and wait until an map is provided and a initial position is specified.

One thing that you may be curious about is that the startup time is slow relative to Python. This is because Julia is a just-in-time or JIT compiled language. Compilation is not performed until a piece of code (i.e. a function) is needed. Note that if you want to hack at this codebase there are ways to get around this (i.e. Revise.jl). Additionally, you can change the optimization level from -O3 to -O0 under the args section of ParticleFilter.launch, although this will negatively affect runtime performance.

Configuration

All settings are located in the scripts/settings.jl file as key-value pairs. You can change the settings here, or via the ROS parameter server. To see which parameters are configurable via the parameter server, you can use rosparam list or rosparam dump after launching the node. Additionally, the node will log to /rosout and the terminal all the settings which are configurable via the parameter server, whether a parameter was found on the server, and what its value is as follows:

 [INFO] [1549348704.911341]: Param "steering2servo_gain" not set. Using default value: "-1.2135"

Future versions will have these parameters documented in a ROS config file.

NOTE: The one exception is multi-threading which is not configurable at runtime. Instead, set the environment variable JULIA_NUM_THREADS to the integer number of threads you would like to use. The default is either the number of physical CPU cores on your machine, or one (I've seen both :)).

To aid in development, please kindly file a git issue if you run across any problems or want something added. Happy localizing!

API

Parameters can be changed in settings.yaml

Publishers

Topic	Type	Description
/pf/inferred_pose	geometry_msgs/PoseStamped	Particle filter pose estimate
/pf/viz/particles	geometry_msgs/PoseArray	Partilcle array. Good for debugging
/pf/viz/laserpose	geometry_msgs/PoseArray	Pose fo the laser
/odom	nav_msgs/Odometry	Odom of car estimate from particle filter

Subscribers

Topic	Type	Description
/scan	sensor_msgs/LaserScan	Current laserscan
/vesc/sensors/servo_position_command	std_msgs/Float64	Current steering angle
/vesc/sensors/core	vesc_msgs/VescStateStamped	Current speed