Improved Socialtaxis for information-theoretic source search using cooperative multiple agents in turbulent environments

Abstract

This paper proposes an improved version of the Socialtaxis approach for efficient source search and accurate estimation in a turbulent atmospheric environment using multiple cooperative agents. The original Socialtaxis, one of the decentralized information-theoretic source search methods for multiple agents, maximizes individual entropy reduction as well as group information diversity for rapid and efficient source search. However, it has several drawbacks: high computational complexity due to the grid-based environment, bias towards exploration, the lack of cooperation among agents (e.g., not sharing measurements and decisions), and the assumption of network connectivity between agents at all times. In order to address the above issues, we improve Socialtaxis in terms of the following aspects. First, we modify the grid representation of the environment of Socialtaxis to the continuous domain using the Rao-Blackwellized particle filter to reduce computational loads and enable more efficient Bayesian estimation of the source parameters. Second, we introduce a new utility function that balances the exploration and exploitation better by using a distance to the estimated source. Third, the sequential greedy algorithm is applied while sharing measurements with one another to realize a fully decentralized decision-making system. Lastly, to prevent network disconnection among agents, we ensure connectivity preservation in a decentralized way so that the agents are within the communication range for information sharing. Extensive numerical simulation results show that the proposed improved Socialtaxis outperforms the original Socialtaxis as well as other existing state-of-the-art source search strategies. The considered aspects of the improved Socialtaxis are proven to be crucial elements of decentralized information-theoretic source search, which can be robust to different wind conditions and encourage multiple agents to cooperate more.