Abstract: To solve the universality and scale controllability issue of the driving authority arbitration in shared control, an analysis and evaluation framework of the driving capability mechanism in the virtual random vehicle-road field(RVRF) is proposed. As the assessment environment for the mechanism analysis of the driving capability, driving scenarios those can reveal the human-vehicle-road coupling mechanism and vehicle-road cooperative rules by coupling the space topology of drivable area with the vehicle motion patterns are obtained in the RVRF. Basing on the strict definition of driving capability, the off-line identification model satisfies the high-order nonlinear attributes of driving capability is established in the RVRF evaluation framework. The sample set consisting of the decoupled and dimensionless key parameters in the model is the key basis to represent the intrinsic attribute of driving capability. The classification is done basing on the combination of subjective and objective and driving capability is identified in real time basing on Gaussian multi-dimension hidden markov model(GM-HMM). Results show that the proposed virtual RVRF can objectively reflect the wave-particle duality of traffic flow and the proposed evaluation method for driving capability in the virtual RVRF can achieve accurate and reliable evaluation results. The mechanism analysis of driving capability in the virtual RVRF optimizes the rationality of driving authority arbitration mechanism, and improves the safety and driver acceptability for the shared control system.

Key words: vehicle engineering, random vehicle-road field, driving capability, machine learning, system identification