Skip to main content



Research Projects

Cochlear Filter

We investigated the possibility of implementing online learning rules in analog VLSI. Specifically, we intend to develop an aVLSI circuit design for a trainable adaptive filter for audio processing that feeds the output of one of Andre’s cochleas into a simple neural network that is adapted on-chip by Nici’s stochastic meta-descent (SMD) rule, reformulated as three coupled continuous-time differential equations Full Project (PDF)

Smart Vision Chip Vision-based navigation (e.g. course stabilization, obstacle avoidance) of indoor flying robots (e.g. blimps, planes) requires real-time image processing capabilities. However, such robots have a dramatically limited available payload (roughly between 10 g for planes and 100 g for blimps). On-board vision modules therefore have tight constraints regarding weight, size and processing speed. Single-chip analog VLSI vision modules with parallel processing can be tailored to meet these requirements for such applications, but the standard packages for VLSI chips, the circuit boards with the biasing and input/output connections, and the optical systems are too bulky and heavy. Full Project (PDF)

WormBot

In the WormBot project we aim to demonstrate elegant robust robotic motion based on simple biologically plausible design principles in a high degree-of-freedom (DOF) system. We investigate in motion generated by multiple 1 DOF segments that are individually controlled by local Central Pattern Generators (CPG), but achieve overall motion stability by short- and longrange coupling. Full Project (PDF) Accompanying Project Video

 

Completed Projects funded by INE

Automated Event Detection in Underwater Video

We demonstrate an attentional selection system for processing video streams from remotely operated underwater vehicles (ROVs) and other underwater platforms such as autonomous underwater vehicles (AUVs) and stationary Ocean Observatories. This system begins to automate the annotation of videotape data, currently a human task. Our approach is to develop and apply neurobiology and neuromorphic information technology to automatically process video from cameras deployed in the ocean. The system identifies potentially interesting visual events, such as organisms entering the field of view, based on low -level spatial properties of salient tokens which are associated with those events and tracked over time. If video contains objects that define such events, they are labeled "interesting", otherwise they are labeled "boring". By marking the interesting events in the output stream, we augment the productivity of human video analysts. Full Project (PDF)