Data fusion with flexible message composition in Driver-in-the-Loop vehicular CPS

This work addresses a unique data fusion problem in Vehicular Cyber-Physical Systems (VCPS) arising from Human Factors (HF) considerations. Typically, a VCPS message intended for human drivers is composed of many data elements (DEs), and different messages can be fused by the sender before transmission e.g., by eliminating identical (or redundant) DEs in order to save transmission bandwidth in the wireless network. Still, not all distinct DEs can be received properly due to the limited transmission resources available to the sender and/or transmission errors. Subsequently, some of the messages intended for a driver cannot be delivered. On the other hand, a partially delivered message may still be beneficial (in terms of generating some utility) to a driver. More specifically, when considering HF, the DEs can be grouped into two distinct parts: essential and auxiliary. While a partially reconstructed message missing even a single essential DE fails to produce any benefit (or utility) for a driver, each auxiliary DE can independently produce an additional utility so long as all the essential DEs of the message are also available. In this paper, we deal with a new Driver-in-the-Loop Data Fusion Problem (DDFP) with the primary issue being: given a list of out-going messages and a limit on the number of DEs that can be transmitted, how does the sender choose which DEs (each carrying a different utility) to transmit, in order to maximize the system-wide utility at the receiver. We formulate DDFP mathematically, and prove it to be NP-Complete. We study DDFP in both ideal and lossy communication networks, and propose several efficient algorithms for them. Besides the Single-Sender-Single-Receiver model, we also look into DDFP in Multiple-Sender-Single-Receiver and Single-Sender-Multiple-Receiver models with several practical considerations. Numerical results from large scale simulations are also presented