Adaptive Sampling using POMDPs with Domain-Specific Considerations

This code is for the paper G. Salhotra*, C. E. Denniston*, D. A. Caron and G. S. Sukhatme, "Adaptive Sampling using POMDPs with Domain-Specific Considerations" IEEE International Conference on Robotics and Automation (ICRA), in press

*Equal contribution

Steps

Prerequisites: pipenv

1. Clone this repo
2. cd /path/to/cloned/repo
3. pipenv install #it will install from Pipfile
4. pipenv shell #enter your new pipenv environment
5. python create_cache.py
6. python <script_name.py>

Run the following files to generate experimental results and plot them. Repository does not include the validation environment data so only analytical functions can be recreated.

File	Description
beta_curve_grid_search.py	Run Experiment to generate data for Figure 2 (Comparing different rollout allocator curves)
figure_makers/beta_curve_test_plot.py	Plot results from beta_curve_grid_search.py (The best rollout allocation curves)
figure_makers/test_plot_beta_distribution	Plot the beta curves the grid search is tested over.
allocator_test.py	Run experiment to generate data for Figure 3 a b & c, (Compare different allocators)
figure_makers/plot_allocator_test.py	Plot the results of allocator_test.py
launch_plan_commit.py	Run experiment to generate data for Figure 3 d & e. (Compare different plan commitment algorithms)
figure_makers/plot_plan_commit.py	Plot results of launch_plan_commit.py
launch_combo_expt.py	Run experiment to generate data for Figure 4 (Compare baseline versus all improvements)
figure_makers/plot_combo_expt.py	Plot the results of launch_combo_expt.py
figure_makers/plot_environment.py	Plot the environments used in the experiments for viewing

Troubleshooting

• if you have dependency issues when creating the envrionment, try to set torch="*" in the Pipfile
• if nothing runs, or if the experiment runner quits immediately, you may have results cached from a previous experiment run. To re-run the experiment, you have to delete (or backup) those results.

To delete cached results, use either of these options

• Move to trash can (Ubuntu 18.04)

mv experiment_runs/<expt_name>/* ~/.local/share/Trash/files
mv smallab.* ~/.local/share/Trash/files

OR:

• Run rm -rf experiment_runs/<expt_name>/* smallab.* to permanently delete the results.

Acknowledgements:

POMCP python implementation from here.