Dataset for Experiments

This repository contain the dataset and the POMDP policies used for the experiments presented in "Clock-state olfactory search in turbulent flows using Q-learning: The geometry of plume recovery". It is a downscaled version of the nose_data.mat dataset available on Zenodo at the following link:

https://zenodo.org/records/6538177#.Yqrl_5BByJE

The dataset provided here corresponds to a spatially downscaled version of the original concentration field, where each spatial dimension is reduced by a factor of 10. The downscaling is performed using the OpenCV cv2 library, allowing the production of a coarse-grained representation of the original field.

Odor source is placed in position y=13 and x = 0 and the data are stored in the format (y-axis, x-axis, time slice).

The simulation environments described in the paper are generated by applying thresholding to the this dataset, setting all values below a specified cutoff to zero. Different plume density regimes are defined as follows:

• Denser: threshold  = 1e−6
• Dense: threshold    = 3e−6
• Sparse: threshold   = 5e−6
• Sparser: threshold  = 7e−6

Code used to perform the experiments can be found at the following repository: https://github.com/PimLb/pynav

POMDP policies are contained in the file "pomdp_policies.zip". This archive contains four files named "policy_gamma_0.98_threshold_<s_thr>_time_4000.pkl", where "<s_thr>" corresponds to the threshold associated with the environment used (i.e., 1e-6, 3e-6, 5e-6, and 7e-6).