Artificial Potential Fields

1) Introduction

You can find the basic implementation of artificial potential fields path planning algorithm. You can try hyperparameters with rqt_reconfigure node in order to see the difference of them on the fly. RViz is used for visualization.

2) Project Packages

base_planner
artificial_potential_field_planner

3) Dependencies

dynamic_reconfigure
geometry_msgs
nav_msgs
gtest
rqt_reconfigure
rviz
tf
map_server

packages are used in this project.

3) Tests

Unit tests

There two different test scenarios. One of them has empty map and the goal is reach the goal point with the max distance of resolution of the map(which is one). Second one is almost the same but there is a big obstacle in the middle of the map. It should get pass through it to the goal with the distance of, again, resolution.

To run unit tests, you should first compile it with;

catkin build --catkin-make-args run_tests

After that, you run tests with;

rosrun artificial_potential_field_planner artificial_potential_field_planner_test

Real-time testing

To run real-time tests, you should first compile it with;

catkin build

There are three maps in the project for real time testing. For these three maps, there also three launch files. In order to create a fixed frame for both map and planner algorithm, static_transform_publisher node of the tf package is used. Origin of the maps are different, so transforms should be too. You can start each test with

roslaunch artificial_potential_field_planner apf_dummy1.launch
roslaunch artificial_potential_field_planner apf_dummy2.launch
roslaunch artificial_potential_field_planner apf_dummy3.launch

commands.

Rviz and rqt_reconfigure panel shall be opened. You can install default algorithm parameters with rqt_reconfigure load button. There is prepared parameter set;

ws/src/artificial_potential_field_planner/params/parameters.yaml

Or, you can try all of the possible values by changing it from the panel.

To see the path planning algorithm working, you should first select a pose estimation by RViz. Then, you can set the goal pose("2D Nav Goal" button). If force field width and height is not zero, a force map is going to be drawn around the last reached point.

Design

All code is written in C++. ROS Noetic is used for this project.

Algorithm implementation is isolated from Ros Infrastructure except for nav_msgs::OccupancyGrid. For other pose or vector representation needs make me create a class called Vector2D(Although it would be better use a third party library like Eigen, that dependency may bring so much unnecessary classes or packages. Size of the algorithm package would be so much). I only use this data structure in this project.

APFPlanner inherits a base planning class called BasePlanner so that different algorithms can be called similarly. BasePlanner has three functions

virtual std::vector<Vector2D> plan(const Vector2D& goal_pose)
virtual void setCurrentPose(const Vector2D& start_pose)
virtual void setMap(nav_msgs::OccupancyGrid map)

These functions are common for all navigation algorithms as far as I know. This is the reason these functions are in BasePlanner.

APFPlanner only have fields and functions about artificial potential fields path planner algorithm.

And also, since some of the classes can be same with the classes of different libraries(like Vector2D), I added namespace for them.

Zdctryto/APF-Path-Planner