Multi-Resolution Model and Context Aware Information Networking for Cooperative Vehicle Efficiency and Safety Systems

Large scale deployment of connected and automated vehicles is impeded by significant technical and scientific gaps, especially when it comes to achieving real-time and high accuracy situational awareness for cooperating vehicles. This CAREER project aims at closing these gaps through developing fundamental information networking methodologies for coordinated control of automated systems. These methodologies are based on the innovative concept of modeled knowledge propagation. The approach is to utilize the novel concepts of model communication and its derived multi-resolution networking. Model-based communication relies on transmission and synchronization of models (e.g., stochastic hybrid system structures and parameters) instead of raw measurements. This allows for high fidelity synchronization of dynamical models of cooperating agents over a network. The models are learnt and updated in real-time as the behavior of the agent evolves. We show examples of how vehicle behavior (or vehicle+driver behavior for manually driven cars) is modeled as a Gaussian Process. The model can be used to track movement of a vehicle with an order of magnitude better accuracy. Another possible form is a simpler ARX-HMM form (Autoregressive exogenous models embedded in an HMM) which has been applied to a Cooperative Adaptive Cruise Control (CACC) application. The results show an order of magnitude improvement in spacing error (disturbance) of CACC. The modeling approaches developed in this project also create the potential of abstracting vehicle+driver behavior as short sequences of mode transitions, which will enable longer term maneuver predictions (in addition to short term movement tracking). This possibility opens the door to applications other than creation of rich situational awareness as it was initially intended.