K-DUMBs IoRT: Knowledge Driven Unified Model Block Sharing in the Internet of Robotic Things

6G is expected to revolutionize the Internet of things (IoT) applications toward a future of completely intelligent and autonomous systems. Conventional machine-learning approaches involve centralizing training data in a data center, where the algorithms can be used for data analysis and inference. To promote green computing in IoT applications, Machine-2-Machine (M2M) technologies are largely focused on lowering energy consumption and creating effective IT infrastructure. In this paper, we introduce an AI-enabled One-Shot Interference(O-SI) Knowledge-Driven unified model block sharing (K-Dumbs) framework in which actionable knowledge is aggregated from the training perception robots to facilitate others at the Edge in the vicinity. To demonstrate the practicality of the proposed concept, we explore a K-Dumb Fed-Average (FedAvg) algorithm to meet the massively distributed and unbalanced pattern and privacy requirement of the Internet of Robotic Things(IoRT). Simulation results show that, when compared to traditional Federated Learning (FL) systems, the proposed K-Dumb FedAvg architecture delivers higher information-sharing and learning quality. In addition, we validate our method using MNIST handwritten digits for training image processing with an accuracy that is close to the centralized solution for up to 80% reduction in the amount of exchange data with the O-SI method. Furthermore, the suggested solution reduces IoRT energy consumption by up to 10 times and protects privacy.