World Modeling for Mobile Platforms Using a Contextual Object-Based Representation of the Environment

Today's mobile robotic systems are expected to handle a great variety of different tasks while reliably operating in highly complex and dynamic environments. This requires a correct and interpretable representation of the robot's surroundings. In recent years, a number of world representations have been developed, tailored for specific fields of application, but with very poor generalization possibilities. In this paper, we close the gap between application-specific world models by introducing a more holistic approach to environment modeling in the form of a Contextual Object-based Representation of the Environment. It is composed of a set of state vectors combining object semantics and the key concepts of the different world modeling techniques into a high-level object-oriented representation. The structure enables the probabilistic modeling of complex and dynamic environments together with all associated uncertainties. By providing the possibility to derive well-established feature-and grid-based world representations at any point in time, we offer ease of integration into existing systems. To demonstrate the aptitude and generalization capabilities of our approach, its application is demonstrated on simulated data. Our approach clearly extends the capabilities of established models by being able to represent both static and dynamic scenarios with semantic object identities in a form suited for navigation, object tracking and interaction.