Situated behavior planning and prediction for guidance and assistance of human operators

The need of human-machine systems is increasing. The supervision of the human behavior allows the detection of missing errors and the support of the human by the decision making. This thesis presents behavior planning methods integrated in a human support decision making system. The supervision of human-machine interaction requires the modeling of human actions and the changes in the environment. The monitoring of the human behavior and the detection of changes in the environment contribute to the safety of human machine systems. Previous works define autonomy levels in relation to human roles. Effects of human factors on the human-machine interaction are analyzed. Tree-based state spaces are developed to predict the human intention and to calculate static action sequences consisting of states and actions. However, the illustration of the human behavior as action and situations based on knowledge experts and the planning of situated action spaces allowing to support the human by the decision making are hardly discussed. In this work, a Situation-Operator modeling (SOM) approach is used to illustrate the human behavior as action sequences consisting of actions modeled as operators and scenes modeled as situation vectors. The SOM-based illustration of the human-machine interaction is applied to real driving scenarios in inland and automotive fields. An automated supervision control is developed to detect warnings and human errors, to generate alternative decision options to avoid critical situations, and to suggest the optimal operation to the human operator. An event-discrete supervisory control loop allows the checking of assumptions and goal reachability, the detection of warnings, and the calculation of resulting situations if the operator is admissible. A strategy is developed for the determination of the optimal decision option based on a modified fuzzy-based cognitive reliability and error analysis method and by considering the goal reachability and the operation time. The event-discrete calculation of possible operators and resulting situation is integrated in a graph-based planning process of decision options according to the SOM-notation. The event-discrete behavior planning can calculate action spaces consisting of possible action sequences leading to the final desired situation. An event-discrete action space-based behavior planning and prediction algorithm is proposed and applied to real driving scenarios in inland and automotive fields. An additional value of this thesis is the situated and event-discrete behavior planning and prediction. Planned decision options and action spaces are situated and can be calculated and updated if new changes in the environment are detected. Furthermore, the developed algorithms can be used for several application fields by defining field specific assumptions, performance conditions, and models.