Graphical Networks and Motion Detection

This works seeks to address fundamental research questions regarding the perception of autonomous vehicles. Most critical to the system is that the system be able to classify, predict and interpret spatial and temporal data. Further, this must be done on a time scale relevant to at least twice the speed of operational speeds of a vehicles to be able to successfully navigate potential head on collisions with other vehicles. Traditional tech requires a rethink, and that’s to use ESN and RC type compute systems as they offer a much more efficient means of processing, training and adaptability over conventional networks. Further, a subset of these systems, graphical networks, work by embedding high dimensional information into a latent space for memorization, retrieval and other things. This ability makes graph nets a prime candidate. We demonstrate the first steps in a deployable graphical network for unmanned vehicles.