Abstract
Autonomous vehicles require highly sophisticated decision-making to determinetheir motion. This paper describes how such functionality can be achieved witha practical rule engine learned from expert driving decisions. We propose analgorithm to create and maintain a rule-based behaviour planner, using atwo-layer rule-based theory. The first layer determines a set of feasibleparametrized behaviours, given the perceived state of the environment. Fromthese, a resolution function chooses the most conservative high-level maneuver.The second layer then reconciles the parameters into a single behaviour. Todemonstrate the practicality of our approach, we report results of itsimplementation in a level-3 autonomous vehicle and its field test in an urbanenvironment.