Uncertainty modeling and runtime verification for autonomous vehicles driving control: A machine learning-based approach

Intelligent Transportation Systems (ITS) are attracting much attention from the industry, academia, and government in staging the new generation of transportation. In the coming years, the human-driven vehicles and autonomous vehicles would co-exist for a long time in uncertain environments. How to efficiently control the autonomous vehicle and improve the interaction accuracy as well as the human drivers’ safety is a hot topic for the autonomous industry. The safety-critical nature of the ITSs demands the system designers to provide provably correct guarantees about the actions, models, control, and performance. To model and recognize the drivers’ behavior, we use machine learning classification algorithms based on the data we get from the uncertain environments. We define a parameterized modeling language stohChart(p) (parameterized stochastic hybrid statecharts) to describe the interactions of agents in ITSs. The learning result of the driver behavior classification is transferred to stohChart(p) as the parameters timely. Then we propose a mapping algorithm to transform stohChart(p) to NPTA (Networks of Probabilistic Timed Automata) and use the statistical model checker UPPAAL-SMC to verify the quantitative properties. So the run-time verification method can help autonomous vehicles make “more intelligent” decisions at run-time. We illustrate our approach by modeling and analyzing a scenario of the autonomous vehicle try to change to a lane occupied by a human-driven car.