Status: Active Development
LEMP (Learning-Enabled Motion Planning) is a light-weight framework that combines the power of machine learning with traditional motion planning techniques. With a focus on fast iteration, LEMP provides a rapid and agile solution solution for developing learning algorithms for motion planning tasks.
$ conda create -n lemp python=3.8
$ conda activate lemp
$ conda install -c conda-forge jupyterlab numpy matplotlib
$ pip install pybullet Pillow scipy
# install torch following the instructions from the pytorch website, for example:
$ conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia
# install torch-geometric following the instructions from the torch-geometric website, for example:
$ pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.1.0+cu118.html
$ pip install torch_geometriccd data
unzip static.zip
unzip dynamic.zipWe provide a bunch of useful notebooks in examples.
| Notebook | Description |
|---|---|
| bit_star_planner.ipynb | Example of the BIT* algorithm for planning. |
| dataset.ipynb | Saving and loading dataset for static obstacles. |
| dataset_dynamic.ipynb | Saving and loading Dataset for dynamic obstacles. |
| dynamic_gnn_planner.ipynb | Integration of GNN models with a dynamic planner. |
| grouping_robot.ipynb | Grouping multiple robot arms as one robot to plan |
| load_environment.ipynb | Visualization of trajectories in environments. |
| load_object.ipynb | Load objects / obstacles to the environment. |
| load_robot.ipynb | Load robot to the environment. |
| object_follow_trajectory.ipynb | Trajectory visualization for objects. |
| robot_follow_trajectory.ipynb | Trajectory visualization for robots. |
| rrt_star_planner.ipynb | Example of the RRT* algorithm for planning. |
| sipp_planner.ipynb | Example of the SIPP* algorithm for dynamic planning. |
| static_gnn_planner.ipynb | Integration of GNN models with a static planner. |