Adaptive Sensing for Terrain Aided Navigation

This paper demonstrates experiments for performing adaptive terrain aided navigation in the context of autonomous underwater vehicles (AUVs) equipped with sonar. The experiments were conducted using a 675 kHz sector scan sonar mounted on a planar robotic positioning system in a 3.0 by 9.0 by 1.0 meter testing tank, enabling controlled and repeatable scenarios. The objective of the adaptive stochastic mapping algorithm is to enable feature-based terrain aided navigation of AUVs in environments where no a priori map is available. The approach assumes that distinctive, point-like features can be extracted from vehicle sensor data. A dead-reckoning error model is incorporated to simulate an AUV's navigation system error growth. An adaptation step based on maximizing the Fisher information gained by the next action of the sensor is coupled with the stochastic mapping algorithm to yield more precise position estimates for features in the environment and the vehicle.