Readme

This codebase serves as the experimentation code for "Constrained Hierarchical Monte Carlo Belief-State Planning" which will be presented at the 2024 IEEE International Conference on Robotics and Automation (ICRA). If you find this repository useful, please cite our work:

@inproceedings{jamgochian2024constrained,
  author = {Arec Jamgochian and Hugo Buurmeijer and Kyle H. Wray and Anthony Corso and Mykel J. Kochenderfer},
  title = {Constrained Hierarchical {M}onte {C}arlo Belief-State Planning},
  booktitle = {IEEE International Conference on Robotics and Automation (ICRA)},
  year = {2024}
}

Installation Instructions

Experiments were run with Julia 1.9.0.

1. Install all recursive repository dependencies using git submodules

   • With, for example, git clone --recurse-submodules https://github.com/sisl/COBTSExperiments.git

2. Install packages for development within a local environment with julia develop.jl

   • This will develop all non-registered packages, even if not changing them. This includes git submodule packages saved locally, as well as problem packages that default install to ~/.julia/dev/. This is necessary to avoid "expected package X to be registered" errors when building from Project.toml...

3. Install remaining packages from Project.toml with ] activate . and ] instantiate.

Experiment Instructions

You can run experiments in locally installed environment with, for example, julia --project=. experiments/run_lightdark.jl. The main experiments for the LightDark, Spillpoint, Bumper Roomba, and Lidar Roomba can be run by executing:

• experiments/run_lightdark.jl
• experiments/run_spillpoint.jl
• experiments/run_bumper_roomba.jl
• experiments/run_lidar_roomba.jl

The experiments to sweep additional options and over search queries can be run by executing experiments/sweep_options_lightdark.jl and experiments/sweep_queries.jl, after which plots can be generated by executing experiments/plot_sweep_options.jl and experiments/plot_sweep_queries.jl.

All of the resulting data that were used in our paper can be seen in icra_results, and can be quickly loaded and summarized, with, for example, julia --project=. 'using COBTSExperiments; load_and_print("icra_results/spillpoint_cobts_10sims.jld2")'.