TY  - JOUR
AB  - Autonomous mobile robots have recently become a popular solution for automating cleaning tasks. In one application, the robot cleans a floor space by traversing and covering it completely. While fulfilling its task, such a robot may create a map of its surroundings. For domestic indoor environments, these maps often consist of rooms connected by passageways. Segmenting the map into these rooms has several uses, such as hierarchical planning of cleaning runs by the robot, or the definition of cleaning plans by the user. Especially in the latter application, the robot-generated room segmentation should match the human understanding of rooms. Here, we present a novel method that solves this problem for the graph of a topo-metric map: first, a classifier identifies those graph edges that cross a border between rooms. This classifier utilizes data from multiple robot sensors, such as obstacle measurements and camera images. Next, we attempt to segment the map at these room–border edges using graph clustering. By training the classifier on user-annotated data, this produces a human-like room segmentation. We optimize and test our method on numerous realistic maps generated by our cleaning-robot prototype and its simulated version. Overall, we find that our method produces more human-like room segmentations compared to mere graph clustering. However, unusual room borders that differ from the training data remain a challeng
DA  - 2017
DO  - 10.3390/robotics6040035
KW  - room segmentation
KW  - domestic cleaning robots
KW  - machine learning
KW  - computer vision
LA  - eng
IS  - 4
PY  - 2017
T2  - Robotics
TI  - Human-Like room segmentation for domestic cleaning robots
UR  - https://nbn-resolving.org/urn:nbn:de:0070-pub-29155209
Y2  - 2024-11-21T22:44:10
ER  -