In order to plan their missions and to carry them outnsuccessfully, mobile robots operating in changing environments need tonkeep track of the state of objects. The perception of changes in thenenvironment and the integration of changes into the robot's world modelnis therefore an important problem in mobile robotics. Most of today'snsystems plan their missions based on static models, thus limiting theirnapplicability. We introduce a method to maintain environment models bynestimating the state of changing objects, e.g. their current positionnand configuration, from sensor data. Unlike other methods, which acquirenand maintain sub-symbolic environment models, our method automaticallynmaintains a symbolic CAD model. The method proposed is a Bayesian statenestimator which computes the maximum likelihood estimate of the state ofna dynamic object by matching templates of the object against proximityninformation obtained by the robot. The algorithm employs Monte CarlonMarkov localization to determine the robot's position in itsnenvironment. The localization provides a probability density of thenrobot's position, and matching takes this density into account, tonachieve robust state estimates even while the robot is moving.nExperiments carried out on a mobile robot in our office environmentnillustrate the capabilities of our approach with respect to thenrobustness of the state estimates
展开▼