Multi-Agent Reinforcement Learning using Continuous Macro-Actions
Hierarchical methods provide a useful structured framework to solve asynchronous rein- forcement learning problems efficiently, they also have the added benefit of transferring knowledge between similar tasks. However most of these approaches have been developed for single-agent environments with discrete action spaces, this limits their potential in real- world settings where most problems require the use of multiple agents working together asynchronously and the action spaces are usually continuous. Hence this project aims to explore the performance of the continuous macro-action algorithms on standard multi- agent reinforcement learning environments that will be extended to the macro-actions framework. The results of this research will provide valuable insights into the capabil- ities of macro-action methods in multi-agent, continuous action space environments and inform future developments in the field.