Shanker Keshavdas and Geert-Jan M. Kruijff


Autonomous robotics, ontology, semantic mapping∗ Language Technology Lab, German Research Centre for Artifi-cial Intelligence, Saarbr¨ucken, Germany; e-mail:∗∗ Nuance Communications, Aachen, Germany; e-mail: gj@ nu-ance.com1 This is a revised and extended version of work previously pub-lished in the proceedings of IASTED International Conferenceon Artificial Intelligence and Applications (AIA 2013).


Our problem is one of a human–robot team exploring a previously unknown disaster scenario together. The team is building up situation awareness, gathering information about the presence and structure of specific objects of interest like victims or threats. For a robot working with a human team, there are several challenges. The robot needs to be efficient at performing it’s tasks due to the time-pressure present in rescue scenarios. At the same time, the robot as a team-member needs to perform these tasks in an apparent manner to the rest of the team. Without that, human users might fail to trust the robot, which can negatively impact overall team performance. In this paper, we present an approach to the field of semantic mapping, as a subset of robotic mapping, aiming to address the problems in both efficiency (task) and apparency (team). First, we assess the situation awareness of rescue workers during a simulated USAR scenario and use this as an empirical basis to build our robot’s spatial model. The approach models the environment from a geometrical– functional viewpoint, establishing where the robot needs to be in an optimal position to gather particular information relative to a 3D-landmark in the environment. The approach combines top-down logical and probabilistic inferences about 3D-structure and robot morphology, with bottom-up quantitative maps. The inferences result in vantage positions for information gathering which are optimal in a quantitative sense (effectivity) and mimic human spatial understanding (apparency). A quantitative evaluation shows that functional mapping leads to significantly better vantage points than a naive approach.

Important Links:

Go Back