Experimental Evaluation and Formal Analysis of High‐Level Tasks with Dynamic Obstacle Anticipation on a Full‐Sized Autonomous Vehicle

Certifying the behavior of autonomous systems is essential to the development and deployment of systems in safety‐critical applications. This paper presents an approach to using a correct‐by‐construction controller with the probabilistic results of dynamic obstacle anticipation, and validates the approach with experimental data obtained from Cornell's full‐scale autonomous vehicle. The obstacle anticipation (used to calculate the probability of collision with dynamic obstacles around the vehicle) is abstracted to a set of Boolean observations, which are then used by the synthesized controller (a state machine generated from temporal logic task specifications). The obstacle anticipation, sensor abstraction, and synthesized controller are implemented on a full‐scale autonomous vehicle, and experimental data are collected and compared with a formal analysis of the probabilistic behavior of the system. A comparison of the results shows good agreement between the formal analysis and the experimental results.