This project focus on quadrotor's Kinodynamic planning frontend, including Kinodynamic-RRT* and Kinodynamic-A*. The word frontend means that it will search a coarse(not optimal) trajectory(not path), and the word kinodynamic means that the trajectory obeys kinamatic constraint(collision free) and dynamic constraint(a differential equation).

If this project benefit you in any way, please don't skimp on your star 🌟 .

The project is based on Paper: Kinodynamic RRT*: Asymptotically optimal motion planning for robots with linear dynamics. And some minor changes:

• There are few impelement of KD tree can do a range search in each dimension, so this project only search neigbhors with position(3 dimension).

• The paper samples full state (in a quadrotor with minimum jerk performance index, full state means 9 dimension: pos, vel, acc), but this project only samples the position of quadrotor. Accordingly, when calculating trajectory we solve partial state OBVP for new sample point and solve full state OBVP for rewriting the tree.

With my test, Kinodynamic-RRT* sample the state space randomly, which it is inefficient. It might be horrible in real-time large scale planning.

Build the project:

catkin_make clean && catkin_make -DCMAKE_BUILD_TYPE=Release -DCMAKE_EXPORT_COMPILE_COMMANDS=1 -j8

Run in terminal:

roslaunch my_simple_planner rviz.launch

another terminal:

roslaunch my_simple_planner test_rrtstar.launch

press G in rviz and select your goal.

Search a trajectory in a maze-like 30m x 20m x 3m environment:

Not start yet, but in TODO list.

I use some source file from repo like mockamap, Teach-Repeat-Replan, kdtree.

Thanks for their high quality code.