Collaborative infotaxis: Searching for a signal-emitting source based on particle filter and Gaussian fitting

To effectively leverage the spatio-temporal sensing capabilities of the team searching for a signal-emitting source, this paper presents a collaborative search method, in which each robot employs the weighted social Bayesian estimation and executes the distributed infotaxis search for the source. Cognition difference between robots, measuring the dissimilarity of probability maps, is specially introduced to obtain the heterogeneous weights of Bayesian estimation. However, the requirement of exchanging the whole probability map presents additional challenges in computation and communication for real-time applications. In this work, a solution for fast low-cost collaborative infotaxis method based on a combination of particle filter and Gaussian fitting is proposed. A particle filter is first employed for the representation of the source probability distribution, which makes the infotaxis strategy computationally tractable for large complex spaces using the limited and tractable amount of randomly drawn particles. By fitting a Gaussian density to the particles, each robot obtains the likelihood weight for social Bayesian estimation by only reporting the mean and the covariance matrix of Gaussian distribution rather than exchanging the whole probability maps. The simulation shows the proposed collaborative infotaxis can achieve an efficient search behavior in complex environments using a small number of particles and a lower communication bandwidth. •Weighted Bayesian estimation implemented by particle filter with a resampling scheme.•Identifying the reliability of the detections from others with cognition differences.•Cognition difference is measured by KL-divergence between two Gaussian densities.•Free energy is substitute for the entropy in the infotaxis decision.