Project realized for the course Self-Organizing Multiagent Systems at Universitat Politècnica de Catalunya.
- Python implementation of the Partaker-Sharer advising framework for multiagent reinforcement learning (PSAF, link to the paper) .
- Algorithm tested in a Predator-Prey scenario, by considering a grid-world environment
- As a first step, we implemented the simple Q-learning algorithm for the single-agent case; we then moved to the case of 2 predators, where we implemented and tested the PSAF framework, proposed in the paper, and we compared it to the case of multiple independent Q-learners
- Learning perfomed in 2 cases: with moving prey and with fixed prey
The folders one_agent and two_agent contain the actual python implementation of the Reinforcement Learning algorithm. More details to run and understand the code are provided in the instructions_readme file.
The report folder contains the pdf of the report and also the presentation slides. The provided report contains a detailed description of the project, with all the performed steps, comparisons and drawn conclusions.