This paper considers the computation of motion strategies to efficiently build polygonal layouts of indoor environments using a mobile robot equipped with a range sensor. This problem requires repeatedly answering the following question while the model is being built: Where should the robot go to perform the next sensing operation? A next-best-view planner is proposed which selects the robot's next position that maximizes the expected amount of new space that will be visible to the sensor. The planner also takes into account matching requirements for reliable self-localization of the robot, as well as physical limitations of the sensor (range, incidence). The paper argues that polygonal layouts are a convenient intermediate model to perform other visual tasks.
You can download the ISRR version of the paper.