Warehouse routing problem solved with q-learning

This repository contains my code for the assignment of the course AE4350 Bio-Inspired Intelligence and Learning for Aerospace Applications. It demonstrates how to solve a warehouse routing problem (related to the TSP or VRP) using q-learning (a reinforcement learning algorithm). Next to the basic version of the problem, I have also added capacitated and multi-agent versions. I have also implemented it in a parallel way (using multiple cores) with Ray. More information about the theory and the results can be found in the report.

Installation

Install using pipenv with Python 3.8.

Usage

• warehouse.py: contains the basic Warehouse class and the basic train function
• warehouse_multiagent.py and warehouse_capacitated.py: contain the multiagent and capacitated versions of the Warehouse class, respectively. The training process is not implemented as a separate function. As these two files now contain a lot of duplicate code, they should in the future be combined. Ideally, there would be only one Warehouse class and one train function containing all features.
• warehouse_parallel.py: contains the Main and Worker classes and the train_parallel function of the parallel q-learning implementation. Warehouse from warehouse.py is reused.
• sensitivity_analyses: contains scripts for sensitivity analyses
• other_results: contains scripts for other results, for instance the comparison between the sequential and the parallel algorithm
• tries: contains script that were used to try out various programming aspects