Many mobile robots rely on a combination of dead-reckoning and GPS for outdoor navigation. Dead-reckoning is prone to inaccuracies and GPS signals can vary and be lost altogether. What's a robot to do? Traditional solutions have focused on localizing from overhead maps or images, which works fine if the robot is a cruise missle but not so well for ground robots. A tree in overhead images is big and green but a ground robot sees a tree trunk. In a new paper, "Cost-based Registration using A Priori Data for Mobile Robot Localization" (PDF format) CMU researchers Ling Xu and Anthony Stentz present a cost-based algorithm that relies on kalman filters and particle filters to produce position estimates. They've field-tested their algorithm on the six-wheeled Crusher robot with some success. See also our earlier article on outdoor navigation research by Anthony Stentz.