The goal of this project is to evolve neuromorphic architectures for adaptive visual navigation of autonomous systems and to revisit conventional assumptions about vision processing in behavioral systems.
A biological and model neuron.
In this work, we will use spiking neural controllers. During the last few years, spiking neurons received an increasing attention from the computer science community. Theoretical hints seem to show that spiking neural networks could be particularly efficient at processing time-dependent sequences. This is a required capability for robot navigation in dynamic environments. In addition to that, spiking neurons are interesting for their biological plausibility, and the possibility of very lightweight implementation in embedded digital hardware.
We will use a model known as Spike Response Model (SRM) which combines rich dynamics and biological plausibility, without being computer intensive. We will also use a lightweight model (called Bits'n'Spikes) that can be easily implemented on microcontrollers.
More on spiking neurons