Self-Supervised Button Recognition for Indoor Mobile Robots

In recent years, with the rapid advances in technology, the precision and accuracy of indoor autonomous driving have also improved remarkably. However, despite the development of these technologies, there are still many difficulties to perform services through real mobile robots. In this paper, we focused on the mobile robot that drives in multiple spaces in a multi-story building using an elevator. In order to operate the elevator, the mobile robot must be able to operate by pressing the elevator button itself. To solve this problem, in this paper, we propose self-supervised button recognition. We created fake buttons for self-supervised learning and placed these on random backgrounds to increase the diversity of the dataset. In addition, the generated realistic buttons are re-generated like real buttons through a GAN. In experiments, we presented that our self-supervised button recognition was performed in an actual environment without separate labeling.