Analog and Digital implementation of retinal ganglion cells for robot navigation systems

The idea behind this PhD, funded by the VISUALISE FP7 project of the European Union which involves four universities, is not only the integration of the Dynamic Vision Sensor in a robot’s navigation system, but also the processing of the essential information extracted, with more strictly biological models, so as to mimic Retinal Ganglion Cells (RGC) in the retina. Applying the computations performed by the retina’s inherent cells, important information can be obtained from the environment quickly, so as to guide the robot more efficiently. The standard and non-standard center-surround ON/OFF response RGC models will be based on the neuron circuits modeled by the partner University of Göttingen. An example of these are the approach cell (a cell detects approaching dark objects) and the object motion cell (a cell that is excited by the motion of objects and inhibited by its self-motion). These type of computations should allow a robot to navigate following or avoiding a particular object for example, and at high speed. As a proof of concept, a practical demonstration will take place with the robots provided by the partner University of Ulster [10] with the simulation of a predator and prey scenario. One robot will be equipped with object motion sensitive RGC to capture the prey and distinguish its movements from its own, the second with an approach RGC to detect and evade the predator. This scenario will be tested under demanding visual conditions and low contrast. If the computation proves to be faster, but also more efficient, than the state of the art in computer vision for robot navigation, this will allow further developments and funding in this area of research.

© 2018 Institut für Neuroinformatik