Spacio-temporal situation assessment for mobile robots

In this paper, we present a framework for situation modeling and assessment for mobile robot applications. We consider situations as data patterns that characterize unique circumstances for the robot, and represented not only by the data but also its temporal and spacial sequence. Dynamic Markov chains are used to model the situation states and sequence, where stream clustering is used for state matching and dealing with noise. In experiments using simulated and real data, we show that we are able to learn a situation sequence for a mobile robot passing through a narrow passage. After learning the situation models we are able to robustly recognize and predict the situation.