Keywords
Deep Learning, Reinforcement Learning, Deep Reinforcement Learning, Artificial Intelligence
Select the category the research project fits.
Mathematical/Computational Sciences
Is this submission part of ICaP/PW (Introductory Composition at Purdue/Professional Writing)?
No
Abstract
The long-standing goal of factory optimization is to find optimal machine and conveyor belt placement to maximize the efficiency of the assembly line. We are developing a reinforcement learning agent to play Factorio, a game where you build and maintain factories, without prior domain knowledge. Factorio is the perfect environment for deep reinforcement learning as it supports extensive modification using an in-game debugging mode which allows our agent to interface with the game effortlessly. The reinforcement learning agent implements a policy of actions based on the reward function, continuously optimizing towards incremental goals specified by the user. The ultimate goal of our agent is to learn how to automate production, find creative solutions to maximize production efficiency in the game, and then transfer this learning to the design and management of real-world factories.
Recommended Citation
Duhan, Shivam; Zhang, Chengming; Jing, Wenyu; and Li, Mingqi, "Factory optimization using deep reinforcement learning AI" (2019). Purdue Undergraduate Research Conference. 57.
https://docs.lib.purdue.edu/purc/2019/Posters/57
Factory optimization using deep reinforcement learning AI
The long-standing goal of factory optimization is to find optimal machine and conveyor belt placement to maximize the efficiency of the assembly line. We are developing a reinforcement learning agent to play Factorio, a game where you build and maintain factories, without prior domain knowledge. Factorio is the perfect environment for deep reinforcement learning as it supports extensive modification using an in-game debugging mode which allows our agent to interface with the game effortlessly. The reinforcement learning agent implements a policy of actions based on the reward function, continuously optimizing towards incremental goals specified by the user. The ultimate goal of our agent is to learn how to automate production, find creative solutions to maximize production efficiency in the game, and then transfer this learning to the design and management of real-world factories.