A custom device for reusable image processing

Some applications are recurrent and could be reused among multiple applications. A typical application is image recognition, i.e. how to map an image from the robot (typically in RGB) to a neuronal network. We assume that the connection to the network is to some degree similar and so we want to create a common denominator to specify such a connection only once and reuse it in multiple applications.

Custom devices provide a way for such a reusable connection.

Custom devices are devices with a custom algorithm to bind to a neuronal network. Unlike other devices that are connected to a neuronal network directly by the brain communication adapter, custom devices provide their own logic on how they connect to a given set of neurons.

Note

We have not really used custom devices so far at this stage of the NRP platform. We will add more documentation here as soon as we have done so.