Abstract: I will discuss sensor-based position estimation for mobile robots. The emphasis will be on the acquisition and integration of stable measurements that allow an observer's pose to computed, and we will take for granted the avilability of suitable filtering schemes that would allow such data to be integrated over time.

The primary theme of the talk will be the computation and use of visual features for pose estimation. Our approach is based in informed sampling of images in the environment to extract robust cues that can be used for pose estimation. Selecting suitable cues depends on the use of a computational interest operator and a subsequent feature recognition operation and interpolation operation.

In the latter part of the talk I will outline a new approach to resolving ambiguity in the pose estimation process. Even for idealized models of sensing and mapping we can show that optimal resolution of ambiguous pose estimates in NP-hard. While a polynomial-time approximating algorithm has been demonstrated (in prior joint work with Romanik and Whitesides), it has certain drawbacks in terms of both cost abnd robustness. A new approach to this difficulty appears to promise better performance.