LANSKY@SRI-AI.ARPA (02/27/86)
REPRESENTATION AND ESTIMATION OF SPATIAL UNCERTAINTY Randy Smith (SMITH@SRI-AI) Robotics Lab, SRI International 11:00 AM, MONDAY, March 3 SRI International, Building E, Room EJ228 (new conference room) Current work on a method for geometrical reasoning under uncertainty will be presented. Such a reasoning component will be important to planning systems for many robotic applications, including autonomous navigation and industrial automation. A general method will be described for estimating the values and estimated errors in the relationship between objects whose locations are represented by coordinate frames. The elements in the relationship may be described by bounding intervals, or may be described by means and covariances, if a statistical model is available. The relationship between the frames (objects) may not be explicitly given, but known only indirectly through a series of spatial relationships, each with its associated error. This estimation method can be used to answer such questions as whether a camera attached to a robot is likely to have a particular object in its field of view. More generally, this method makes it possible to decide in advance if an uncertain relationship is known accurately enough for some task to be accomplished, and if not, how much of an improvement in locational knowledge a proposed sensing action will provide. The calculated estimates agree very well with those from an independent Monte Carlo simulation. The method presented can be generalized to six degrees of freedom, and provides a practical means of estimating the relationships (position and orientation) between objects as well as the uncertainty associated with the relationship. -------