Semantic Knowledge Consensus for Multi-agent Collaboration: A Naming Game Approach

Semantic communication enables collaborative agents in networked systems to communicate and collaborate in an efficient and intelligent manner, by deeply exploiting the computing capacities and knowledge of the agents. However, frequent changes in environments, as well as high uncertainties in agents, such as agent types, movements, and objected tasks, will lead to heterogeneous and asynchronous knowledge bases within different agents, resulting in failed communications and collaborations. To shed light on this issue, we propose and analyze a naming game based mechanism for multi-agent semantic knowledge base consensus. In this approach, each agent extracts necessary semantic features through learning, and interacts with other agents using a simple protocol to reach knowledge consensus through a naming game. Semantic knowledge base consensus is achieved as such. Finally, experimental results demonstrate that the proposed method can effectively improve collaboration of agents in the condition of heterogeneous semantic knowledge.