Abstract
This paper proposes a radiation source exploration system using autonomous mobile robots equipped with heterogeneous sensors for unknown radioactive environments. Conventional exploration faces trade-offs between efficiency, cost, and the occurrence of false peaks, known as ghosts. To address this, we combine a robot equipped with a directional sensor to globally identify potential source candidates (task points) and robots with non-directional sensors to pinpoint the exact locations. Task points are extracted using a von Mises distribution-based heatmap and assigned via a market-based allocation method. Upon reaching the points using D* Lite path planning, the robots estimate the source locations using Gaussian Process Regression (GPR). Simulation results in an environment mimicking the Fukushima Daiichi Nuclear Power Station demonstrate that the proposed method accurately estimates source positions with an average error of 0.18 m, successfully eliminating ghost peaks and compensating for individual sensor disadvantages.
