ABSTRACT. A new verification framework is presented for the decision making of autonomous vehicles (AVs). The overall structure of the framework consists of: (1) A perception system of sensors that feed into a (2) a reasoning agent based on a Jason architecture that operates onboard an AV and interacts with a (3) model of the environment. The agent relies on a set of rules, planners and actions to achieve its ultimate goal of driving the AV safely from a starting point to its destination. The verification framework deploys an innovative combination of two well known verification tools: (1) the model checker for multi-agent systems (MCMAS) to check the consistency and stability of the agent rules before choosing any action and (2) the PRISM probabilistic model checker to provide the agent with the probability of success when it decides to take a planned movement sequence. The agent will select the movement-actions with the highest probability of success. The planned actions are executed by a control system to control the movement of the AV on the ground using a set of primitive movement skills using its actuators. The framework uses the Robot Operating System (ROS) to implement the reasoning agent and the Gazebo Simulator to model the AV, its sensors and the surrounding environment. The agent connects to a MATLAB-based perception and control system to steer the AV.