Björn Jensen

Narrative situation assessment for human-robot interaction

In this paper we address the problem of interpreting sensory data for human-robot interaction, especially when gathered from several robots at the same time. After describing motion tracking in this context, we introduce a general framework for situation representation, and how it simplifies extraction of information suitable for complex machine dialogs. As a concrete implementation thereof, a narrative description of a complex scene in a public exposition is created. We regard issues of interpreting sensor data in an efficient way and discuss the effects of the number of robots on the result of the scene interpretation to show that our approach is not only scalable but also profits from a growing number of robots.

Detecting Semi-static Objects with a Laser Scanner

A method to register dynamic and semi-static objects with an a-priori known static map is proposed. Candidates for semi-static objects are extracted from laser data based on a-priori information (shape and size) of common indoor objects. Expectation Maximization serves at the same time scan alignment and data association between scan data andthe static a-priori map and the semi-static map. Dynamic objects are identified as outliers in data association. Evidence of visible but unmatched parts is gathered using recursive Bayesian updates to yield reliable candidate rejection in the presence of sensor noise. This allows the resulting semi-static map to adapt to changes in the environment, which is demonstrated experimentally.