Efficient high speed signal estimation with neuromorphic vision sensors

摘要

Recently developed neuromorphic vision sensors present a high speed, event-based alternative to conventional vision in robotic systems. We present a method for the design of simple, low computation estimators that take advantage of the remarkable properties of these sensors to provide low latency, high bandwidth sensing for feedback control of fast dynamical systems. It is shown that under certain circumstances a simple transformation of the event stream from such a sensor can allow it to be treated as an asynchronous configuration sensor, with minimal computation required to achieve high speed signal reconstruction. These results are applicable to any robotic control problems requiring high performance visual feedback control with low computation.