Real-Time Pose-Aware Low Observability Path Planning

This is the source code for real-time pose-aware path planning.

Installation

Steps

# Clone the repository
$ git clone https://github.com/mit-acl/low_observability_planning
$ cd low_observability_planning

# Create a virtual environment (optional but recommended)
$ python3.8 -m venv planning_env  # We used python3.8, other versions of python might also work.
$ source planning_env/bin/activate  

# Install the package
$ pip install -e .
# Install required dependencies
$ pip install -r requirements.txt

Running examples with the pretrained policy

A checkpoint is provided in the checkpoint folder. To see an example of how to use it, see notebooks/student_policy_investigation.ipynb, sample data is provided under data_mcmc and data_rrt. Note that the paths in the yaml files need to be changed accordingly (see the next section).

Note that under the notebooks folder there are several other notebooks that investigate different modules.

Collect Data and Train a Policy

First navigate to evasion_guidance folder.

The first step is to collect paths with RRT*, this can be done via the following command:

$ python scripts/collect_rrt_data.py --rrt_config params/rrt_data_collection.yaml

Note that the output_path in the yaml file should be changed to where you want to save the collected data.

Then we need to run MCMC to refine the paths. This can be done via:

$ python scripts/mcmc_refine.py --rrt_config params/rrt_data_collection.yaml --mcmc_config params/mcmc_refine.yaml

In mcmc_refine.yaml, the rrt_data_path should be changed where you saved your rrt data (i.e. output_path in rrt_data_collection.yaml). And the output_path in mcmc_refine.yaml should be changed to where you want to save the refined paths.

To train a student policy, simply run the following command:

$ python scripts/bc_train.py params/bc_train.yaml

Note that the paths in bc_train.yaml should be changed accordingly.