A scalable approach of message interpretation by demonstrations for multi-robot communication

We present an innovative multi-robot communication idea of generic message interpretation system based on updated feed forward network (FFN). A message is passed using demonstration by robotic arm. Recurrent network model RNM is used to learn complex tasks' demonstration then simply learning action sequences; RNM comes with the limitations of time efficiency, storage and provides a rigid structure for saving and retrieving the input data. We propose dynamic message interpretation architecture that is efficient in time and storage. Inefficient recurrent nodes are replaced with updated FFN. This modified architecture is based on hash table. A single hash store is used instead of multiple inefficient context modules of recurrent networks. History for input usability is saved for experience based task learning. We present generalization of this design for multi-robot environments: 1-N (one sender and many receivers) and 1-M-N (one sender, many mediators and many receivers). This system is equally applicable for any kind of robot imitation scenario. Performance evaluation of this approach makes success guarantee for robot message comprehension.