The YASE Framework: Holistic Scenario Modeling with Behavior Trees

Abstract

Intelligent vehicles systems are becoming more complex and with that the development and testing complexity of such systems. As real world tests are not sufficient anymore, simulation driven testing and development becomes indispensably important. A key element of intelligent vehicles simulation is hereby the generation of dynamic traffic situations via scenario engines. Such frameworks need to be flexible enough to model highly diverse concrete and implicit scenarios. To meet these demanding needs, we developed the OpenSource YASE (Yet Agnostic/ Yet Another Scenario Engine) framework. It contains scenario file compiler frontends to load scenarios and a behavior tree implementation for scenario execution. In this paper we present the novel approach of YASE to combine concrete scenarios, implicit traffic, scenery behavior and test criteria evaluation into one holistic framework. This modular approach allows to compose any arbitrary behavior needed for intelligent vehicle development and testing. First, the general functionality of behavior trees is introduced. N ext, we demonstrate the implementation of a concrete Euro NCAP scenario. For this scenario we explain the execution tree traversals for exemplary traffic situations. This is followed by implementations of a driver and a traffic model as a behavior tree. We elaborate in this context the composability characteristics of the used approach by reusing nodes in different contexts and combining scenario nodes with models. Finally, we highlight the model capabilities with a large scaled simulation with more then 8 million covered kilometers of highway traffic within the time frame of 15 simulated years.