Multi-agent communication and behaviour training using reinforcement learning
Simon Vanneste · 2024
There are many real-world problems where distributed systems must work together to achieve a common goal. Artificial intelligence has started to play an important role in our lives. Therefore, we investigated how we can use it to develop these distributed systems. In this research, we explore how different intelligent entities (agents) can work together and communicate with each other. We employ reinforcement learning to allow the agents to learn how to communicate and how to behave. In reinforcement learning, the agent learns which actions it needs to take based on the reward it receives. This training method allows the agents to develop a custom communication protocol that is thoroughly integrated with the trained behaviour. In the first part of the thesis, we investigated how we can use these kinds of methods in real-life applications. Next, we developed multiple algorithms to learn a communication protocol. Finally, we explore how we can train these systems in a decentralized way.