Spike Neuromorphic VLSI-Based Bat Echolocation for Micro-Aerial Vehicle Guidance.

摘要

We summarize the state of the various projects our laboratories have pursued during the course of this support. This includes multiple efforts related to a VLSI-based echolocation system being developed in one of our laboratories from algorithm development, bat flight data analysis, to VLSI circuit design and testing of these algorithms. We have pursued investigations into the spike-based implementation of the interaural-intensity processing regions of the bat brainstem and midbrain with our VLSI modeling of the lateral superior olive (LSO), the dorsal nucleus of the lateral lemniscus (DNLL), and the inferior colliculus (IC). We summarize an exciting new perspective that neurobiologically-realistic, conductance-based synaptic integration is better suited for the particular computations we are seeking. In collaborative work with Cynthia Moss, we have been analyzing previously-captured bat flight data to understand the strategy the bat appears to use in the capture of moving targets. The analysis and computations underlying this problem is closely aligned to the techniques used in describing a control law for explaining a hunting behavior ('motion camouflage') observed in the visually-guided dragonfly. We continue to develop new circuits for an ultrasonic cochlea and have uncovered interesting new issues in our choice for representing the intensity of signals. We have just finished testing the first chip version of an echo-timing-based algorithm ('openspace') for sonar-guided navigation amidst multiple obstacles.