Fault Tolerance Technique Using Bidirectional Hetero-Associative Memory for Self-Reconfigurable Programmable Matter

Abstract

Programmable Matter (PM) based on modular robots is a material which can be reprogrammed to have different shapes and to change its physical properties on demand. It can be deployed in several domains and has a variety of applications in construction, surgery, environmental science, space exploration, etc. PM is composed of a big number of limited resources connected robots called modules or particles to form its shape. These modules communicate with each other and move around each other dynamically in order to switch from one configuration to another. Due to the limited resources of modules and the high number of packets that transit within the system, it is very challenging to ensure packet delivery with high reliability. In this paper, we are using a Bidirectional Hetero-Associative Memory (BHAM) networks to improve the reliability and fault tolerance in PM. The idea is to let modules sending packets with smaller size without loosing any information. Furthermore, this model is also capable to remove noise from received packets. The proposed approach is tested on a real programmable matter blinky blocks platform as well as via simulations. We studied two versions of artificial neural networks based on storage capacity. The experimental results show that the studied approach is efficient in reducing the size of packets that transit in the system thus reducing energy consumption and it is capable to detect and remove noise and correct noisy packets.