How to Evaluate Proving Grounds for Self-Driving? A Quantitative Approach

Proving ground has been a critical component in testing and validation for Connected and Automated Vehicles (CAV). Although quite a few world-class testing facilities have been under construction over the years, the evaluation of proving grounds themselves as testing approaches has rarely been studied. In this paper, we investigate the effectiveness of CAV proving grounds by its capability to recreate real-world traffic scenarios. We extract typical use cases from naturalistic driving events leveraging non-parametric Bayesian learning techniques. Then, we contribute to a generative sample-based optimization approach to assess the compatibility between traffic scenarios and proving ground road structure. We evaluate the effectiveness of our approach with three CAV testing facilities: Mcity, Almono (Uber ATG), and Kcity. Experiments show that our approach is effective in evaluating the capability of a given CAV proving ground to accommodate real-world driving scenarios.