Underwater object tracking using electrosense

Few effective technologies exist for sensing in dark or murky underwater situations, especially in close-range contexts. Yet certain species of aquatic animals are able to navigate effectively under such conditions, using self-generated electric fields. These freshwater fishes are known as "weakly electric fish." They emit a weak electric field (mV/cm near the surface) and sense perturbations of this field induced by objects that differ in conductivity from surrounding fluid, allowing them to navigate and hunt. These animals are known as "active electrolocators." In addition, a number of other species are capable of sensing externally generated fields, including sharks, sturgeon and catfish. These are known as "passive electrolocators." Sharks, for example, may use these cues for navigation as well as for identifying prey. Using these biological systems as working prototypes and inspiration, this thesis explores the use of passive and active electrolocation for velocity estimation, object tracking, and identification of some features of underwater objects. Our goal is to translate biological passive and active electrosense into robotic devices that are guided by these sensory cues. We first explore how we might estimate the traveling velocity of an object moving within water through Earth's field using the shark's passive electrolocation system. We present novel designs based on the principles of electromagnetic induction that are able to estimate velocity and integrate position in a strong artificial field, then in Earth's magnetic field. Next, using weakly electric fish as inspiration, we explore a number of different field emission and sensing configurations and how they can be used for estimating velocity, tracking objects, and identifying features of simple shapes. We explore some novel metrics that relate measurements to physical features and position of holes drilled in sandstone, using a single sensor and emitter pair. We then extend this sensor configuration to a larger number of electrodes and attempt to use this approach to estimate velocity with respect to stationary features in the environment. A simple method of cross-correlating signals between measurements provides a relative measure of traveling velocity. The accuracy of this method varies with distance due to distortions in measured voltages. In order to account for this distortion we consider another, more general, approach used for electrical impedance tomography (EIT) which attempts to solve for the conductivity of a volume by knowing the boundary geometry and electrical conditions. Using this approach with a 45 cm long sensory platform, we demonstrate an ability to estimate object velocity with less than 4% error and position error under 2 cm with distances at or below 20 cm. To the author's knowledge this is the first example of EIT being applied to a mobile platform where electrodes are spread over the surface of a robot, looking outward. In addition this appears to be the first attempt to track an object's position as it moves through a fluid using this technique.