Geo-spatial risk mapping of potential landmine areas using probabilistic estimation approaches on a low-cost mobile robot

Abstract

This work presents the development and evaluation of probabilistic approaches for detecting hazardous zones from buried landmines in outdoor environments. A low-cost mobile robot, equipped with metal detectors and stereo cameras, enables real-time geo-spatial data acquisition and terrain mapping. Three uncertainty modeling techniques are compared: (i) Bayesian spatial mapping with convolution inference, (ii) a diffusion model using partial differential equations, and (iii) geo-statistical kriging. The teleoperated robot generates 2D and 3D probabilistic risk maps from terrain reconstructions via visual localization and mapping. Field experiments validate the effectiveness of these methods in distinguishing safe and unsafe zones, showing that the Bayesian approach achieves the best uncertainty representation and spatial consistency. This work enables scalable, data-driven risk modeling in real-time robotics with applications in demining and hazardous environment exploration.