In this paper we present Robox, a mobile robot designed for autonomous operation in a mass exhibition environment. Robox has
unique multi-modal interaction capabilities and a novel approach to localization using multiple Gaussian hypotheses. What
makes Robox one of a kind is on the one hand its design and the variety of functionalities united in one platform and on the
other hand the scale of the Expo.02 project where Robox has been deployed.
Here, we adopt an experimental view of the task. After the problem specification of mass exhibitions, we outline system integration
aspects: mechanical design, safety, software and hardware architecture and interaction modalities. Finally, seen as an enabling
technology for robots in exhibitions, we present the localization approach in more detail.
Building on former experience with feature-based Kalman filter localization we address the data association problem whose
neglect was found to be the predominant reason for localization failures. Multiple hypotheses are generated by a constraint-based
search in the tree of local-to-global associations, given a local map of observed features and a global map of the environment.
As soon as hypotheses are available they get tracked with an algorithm relying on the same interpretation tree technique.
By track splitting under geometric constraints, location ambiguity can be represented not only globally but also locally,
thus forming a consistent framework for global Kalman filter localization. The experiments demonstrate significantly improved
robustness at modest computational costs.
The raison d’être of Robox is the Robotics pavilion at the Swiss National Exhibition Expo.02. There, ten Roboxes guided more
than half a million visitors through the exhibition, eleven hours per day, seven days per week, from May 15 to October 20,
2002.