Improved Mobile Robot Localization Using Semantic World Models

In mobile robotics, simultaneous localization and mapping is a complex problem. However, by using smart constraints, the problem can be reduced considerably. Instead of constraining the issue to a specific robotic system or its movement behavior, we show how semantic environment perception and modeling allows another point of view and therefore a convenient solution for the problem. We present a method for applicationindependent localization and mapping based on semantic landmarks and the concept of visual odometry. Central starting point is a generic landmark definition, allowing for a reduction of the 3d localization problem to a more efficient search for an affine transformation in 2d space. These semantic landmarks are used to simultaneously map the surrounding environment of the robot, resulting in a widely applicable world model.