View the IROS 2019 submission here and the video on Youtube.
- Python 2.7
- SciPy 1.1.0
- configparser
To train model, run:
python2 train_routing.py
This will generate a file in the model folder with the filename corresponding to grid_filename
in params.cfg
.
A trained model is provided: ./models/grid19.pkl
To train model, run:
python2 train_spacing.py
To test model, run:
python2 test_one.py
This will load the trained model listed for grid_filename
in params.cfg
and display a plot of a generated trajectory.
If you close this window, the program will go on to another test case and show another graph.
The expert trajectory is denoted in green, while the learner's sequence of motion primitives is denoted with red arrows, and a blue spline interpolates these points.
To test model, run:
python2 test_multiple.py
To modify experiment parameters, change cfg/params.cfg
and retrain the model.
- Demo in a simulator
- Enforce trajectory start time and end time - should this change the heuristic?
- Take-offs
- Prune high cost trajectories (as if high cost regions are fake obstacles)
- Refine splines by adding control points sampled using a Variational Autoencoder ( won't generalize to landing vs takeoff)
This project is released under the MIT License. Please review the License file for more details.