Posterior Sampling Reinforcement Learning for Olfactory Search

Project for intership at DSSC

Author: Irene Brugnara
Tutor: Prof. Antonio Celani

The aim of the project is to apply Posterior Sampling Reinforcement Learning algorithm [1] for an olfactory search problem.
The model of odor detections in atmosphere is based on [2].

The code is adapted from https://github.com/IreneBrugnara/RLProject

Source files:

• The file lib/gridworld.py contains the implementation of the abstract base class Gridworld, defining a two-dimensional grid environment in which the search takes place. The method search implements the search algorithm. Classes defined in all other files in lib derive from Gridworld.
• The directory animation contains classes for visualizing search trajectories.
• Files benchmark_1.py and analysis_1.py contain code to respectively collect and process data on large-scale runs of the algorithm (the former produces a pickle file which is to be read by the latter). The benchmark can be run in parallel with multiple cores. Similarly for benchmark_2.py and analysis_2.py.

[1] Osband, I., Russo, D., & Van Roy, B. (2013). (More) efficient reinforcement learning via posterior sampling. arXiv preprint arXiv:1306.0940.
[2] Celani, A., Villermaux, E., & Vergassola, M. (2014). Odor landscapes in turbulent environments. Physical Review X, 4(4), 041015.