Event

Microscopic ground texture enables micron-level robot localization

1 reports · 1 sources

Aaron Wilhelm et al. posted a paper on arXiv proposing a global localization system for mobile robots that uses microscopic optical observation of natural ground microtexture; across multiple surface textures, all localization attempts succeeded in each of 1,007 held-out queries per texture, with translational RMSE of 28–73 micrometers. The system also achieved localization on a low-cost mobile robot using maps collected with a calibrated precision gantry, and can resolve micron-scale displacements caused by thermal expansion; the difference between independent photometric registration and feature-based localization estimates is approximately 9 micrometers RMSE. The authors established the relationship between optical magnification and image-space localization uncertainty with physical localization accuracy, and addressed calibration, camera modeling, and focusing issues in high-magnification imaging.