Landmine detection using an autonomous terrain-scanning robot

This paper describes the software of a terrain scanning robot capable of autonomously manipulating a typical handheld detector for remote sensing of buried landmines in a manner similar to a human operator. The autonomous manipulation of the detector on unknown terrain requires an online terrain map to generate an obstacle free path for the end effector of the robot. The software includes a twofold process of map building and path planning that is implemented into a real-time software platform to take place in parallel to the other functions of the robot. Map building features a distributed sensor fusion system to tackle the uncertainties associated with the sensor data. It provides local terrain maps by fusing the redundant measurements and complementary data obtained from competitive rangefinders and joint position sensors, respectively. The fusion takes place in a multi-step data processing module that includes a batch processing filter, a static filter, and a fuzzy adaptive Kalman filter. The latter requires the dynamic model of the process so that a stochastic model is introduced for the terrain undulations. An important parameter of the model, which significantly influences the output of the filter, is the standard deviation of the probability distribution of the process disturbances. A systematic fuzzy modeling technique is used to determine the standard deviation based on the terrain type and to adapt the filter, accordingly. The outlier rejection is carried out using the Mahalanobis distance between the estimated states of the system and the new measurements. Path planning is carried out based on the terrain map to move the detector at a constant distance and parallel to the ground. Unlike the traditional methods, the path is generated in the non-Cartesian coordinate frame of the sensors to avoid a great deal of transformations involved in reproducing the terrain map in a Cartesian coordinate frame.