Reinforcement learning for swarm systems

• Main Author: Arumugam, Ramaswamy
• Other Authors: Zinovi Rabinovich
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/163384
• Institution: Nanyang Technological University

The application of deep reinforcement learning to swarm systems is currently an actively explored topic. Adapting multi-agent reinforcement learning algorithms to swarm systems is difficult because of dynamic neighbourhood sizes and the lack of agent identities. Hence a key component to building a good Swarm RL algorithm is an information summarization module. There is currently no consensus on the best way to summarize the information from an agent's neighbourhood. Therefore we explore various techniques for information summarization. We evaluate these techniques on two tasks - cover and cluster. We also introduce a new method for summarization based on selecting the top K most important pieces of information from an agent's observation. In this paper, we provide an experimental study of our algorithm and its efficacy.