Modern autonomous systems are performing complex tasks in a real-world environment. This requires a comprehensive overview on the environment, for which the mere storage and retrieval of acquired information is not sufficient. Sophisticated cognitive processing like situation recognition or proactive planning can be realized on the basis of consistent and efficient world modeling. Since autonomous systems have to cope with uncertain and incomplete information, probabilistic information management mechanisms are additionally required.
This contribution introduces new information management mechanisms for environment modeling. The proposed system uses a three pillar information architecture consisting of a prior knowledge, world model, and sensor data. The described Bayesian framework formalizes the information management, including information representaton by means of Degree-of-Belief (DoB) distributions with instantiation, deletion, and fusion mechanisms. In this contribution, a special focus is given to observation-to-instance mapping and decision mechanisms for creating a new instance or updating already existing instances in the model.