Bio-inspired underwater electrolocation through adaptive system identification

Electrolocation is a method of sensing and navigating around nearby objects by probing the environment with a series of electrical pulses and measuring the response. This method, found in several species of electric fish, has the potential for faster response times and reduced scanning overheads when compared to traditional underwater location methods such as sonar. This work describes a biology-inspired model and process method for emulating this sensing modality. Previous work in this area uses parametric models, requiring the learning of many time-varying physical parameters. This limits the usability and adaptability of these methods. Instead of relying on complex physical models, we propose in this paper, a dynamic non-parametric model for underwater electrolocation which can be identified using existing system identification techniques. We further describe ways in which results from adaptive filtering and machine learning can be used to process incoming sensory information for electrolocation. We demonstrate the performance of the proposed improvements using an experimental aquatic testbed. Our experiments shows a 3 × increase in the detection range.