Autonomous Robotic Landmine Detection, Mapping And Avoidance

Abstract

The objective of this thesis is to develop detection algorithm to can solve socially challenging issue of Landmine mapping & removal. Landmines which used to be a safety measure during wars, play a dangerous role in civilian life, in and around the post combat regions with approximately 78 countries observing an estimate of 42 human lives killed every day. To prevent the number of casualties and manual mine clearance, mobile robots can be used for detection & mapping of the mine field region.Mapping, detection and avoidance are the primary challenges faced on a landmine removal. In this thesis, mine center is estimated using non-linear optimization and validated through simulations for mapping. Distinguishing between the surrogate mines from the metals is accomplished using Support Vector Machine (SVM) classification, an algorithm which is formulated and validated through simulations. After detection and mapping of mines, a potential field method is employed for avoidance in simulation and incorporated the ROS (Robot Operating System) Navigation Stack on the actual robot which follows selective A* and Dijkstra's algorithm.Results show that the Gaussian field parameter estimation localizes the mine appropriately even if the detected signals are not on the center of the mine and this is achieved through constrained non-linear optimization. This adds a repulsive force around the mine avoiding the step-over with a new navigation path generation. The SVM classifier simulations provide a clear distinguishing base line between the mines and metals based on the three channel data inputted. In future, with further anomalies into consideration, a working autonomous robot would be achieved with detection, mapping and avoidance of landmines.