Summary
2023
Session Number:A4L-1
Session:
Number:A4L-12
Model Based Multi-Agent Reinforcement Learning
Carlucho Ignacio, Vecchio Giuseppe, Palazzo Simone,
pp.150-150
Publication Date:2023-09-21
Online ISSN:2188-5079
DOI:10.34385/proc.76.A4L-12
PDF download (248.7KB)
Summary:
Multi-agent Reinforcement Learning (MARL) has proven to be a powerful approach for training agents to perform a wide range of tasks. However, traditional MARL methods rely on trial-and-error learning, which can be slow, costly, and can cause damage to the robots when applied to real-world systems, especially during learning stages. On the other hand, Model-Based Multi-agent Reinforcement Learning (MBMARL) uses a model of the environment to generate predictions and plan actions, which can significantly speed up the learning process. This model-generated data can reduce the cost of training considerably, which is particularly helpful in robotic applications.