Low Altitude Path Planning

Installation

Steps

Creating a python virtual environment and install dependencies using requirements.txt. Navigate to imitation_learning and install the low_altitude_nav package with

$ pip install -e .

Running examples with the pretrained policy

Pretrained checkpoints with example data and the corresponding terrain is provided here.

To inspect the data and the pretrained policy, see examples in imitation_learning/notebooks.

To train a policy, use:

$ python3 scripts/train_resnet.py params/resnet.yaml

(Please change the paths involved accordingly.)

Some Tips on Isaac Sim

To load a usd file into Isaac, use

omni.usd.get_context().open_stage("terrain.usd")

Camera pose can be set with

camera.set_world_poses(camera_positions, camera_orientations, convention='world')

Related Works

This work builds on:
Safe Low-Altitude Navigation in Steep Terrain with Fixed-Wing Aerial Vehicles
Learning High-Speed Flight in the Wild
iPlanner: Imperative Path Planning
Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics