STITCHER: Constrained Trajectory Planning in Complex Environments with Real-Time Motion Primitive Search

[ArXiv] [Code]

STITCHER is an optimization-free trajectory planning framework that stitches together motion primitives (i.e. short trajectory segments) to compute long-range, expressive, and near-optimal trajectories in real time. STITCHER has provable performance with guaranteed hard constraint satisfaction and a priori time/memory bounds. The framework has been extensively tested in both simulation and hardware on a custom quadrotor.

Dependencies

• Ubuntu 20.04
• ROS Noetic
• C++ 17
• catkin tools
• CMake >= 3.16.3
• Point Cloud Library >= 1.10.0
• Eigen >= 3.3.7
• LibYAML >= 0.6.2

sudo apt install python3-catkin-tools cmake libpcl-dev libeigen3-dev libyaml-cpp-dev

Compiling

Create a catkin workspace, clone the stitcher repository into the src folder, and compile via the catkin_tools package:

mkdir ws && cd ws && mkdir src && catkin init && cd src
git clone https://github.com/vectr-ucla/stitcher.git
catkin build

Execution

After sourcing the catkin workspace, launch the STITCHER planner ros node.

cd ws
source devel/setup.bash
roslaunch stitcher_planner goal_click.launch

If successful, RViz will open with the loaded environment. To generate a path, use the 2D Nav Goal tool in RViz to click a start and goal point.

Test Environments

We include two test environments in data: perlin_map.pcd, and office.pcd (referred to as Willow Garage in the manuscript). Edit goal_click.launch with the environment you would like to test in.

Note that the office environment requires a smaller resolution for the geometric path planner (e.g. A*, JPS) in order to move through doorways, while the perlin map may use a larger resolution for faster computation of the geometric path. The resolution can be changed in cfg/params.yaml.

The following are values that work well.

• Perlin Noise: resolution = 0.5
• Office: resolution = 0.15

Quadrotor Flight

Our framework was tested on hardware and shown to generate dynamically feasible paths. Below, is one experiment of our quadrotor flying a path generated by STITCHER.

Citation

If you found this work useful, please cite our manuscript:

@article{Levy25:STITCHER,
  author={Levy, Helene J. and Lopez, Brett T. },
  journal={ArXiv}, 
  title={STITCHER: Constrained trajectory planning in complex environments with real-time motion primitive search}, 
  year={2025}
}