Micro-Localization: High-Precision Mobile Robot Localization Using Microscopic Ground Texture

Aaron Wilhelm; Marina de Albuquerque Marcon; Emith Uyanwatte; Nils Napp

cs.ROOct 15, 2026

Micro-Localization: High-Precision Mobile Robot Localization Using Microscopic Ground Texture

Authors: Aaron Wilhelm, Marina de Albuquerque Marcon, Emith Uyanwatte, Nils Napp

Organizations: Department of Mechanical Engineering, University of Michigan, Ann Arbor, MI 48109, USA.

Abstract

We demonstrate high-precision global localization using natural ground texture viewed through microscope optics, achieving localization in the tens-of-micrometers regime. For the planar ground texture localization problem considered here, we relate optical magnification and image-space localization uncertainty to physical localization precision. Leveraging this relationship, we then develop a microscope-based localization system that addresses the calibration, camera-modeling, and focus challenges introduced by high-magnification imaging. We evaluate this system across several surface textures using a calibrated precision gantry. Across 1,007 held-out queries per surface, the system localizes every query and achieves translation RMSEs of 28 $-$ 73 $μ$ m. Independent photometric registration agrees with the feature-based localization estimates to within approximately 9 $μ$ m RMSE. We further demonstrate localization on a low-cost mobile robot using the gantry-collected maps, and also show that the system can resolve micrometer-scale displacement caused by thermal expansion. These results show that microscopic natural surface texture can bring mobile localization into a precision regime typically associated with fixed positioning systems.

Figures & tables

Fig. 1 : System Hardware, Multi-Scale Texture, and Localization Performance . (A) Custom mobile platform equipped with the downward-facing camera, microscope optics, and illumination. (B) Macroscopic view of concrete and (C) the corresponding microscopic view, revealing distinctive surface structure used for localization. (D) Spatial localization performance over the mapped workspace, showing gantry query errors and mobile-robot trajectories colored by translational error.

Fig. 2 : Macroscopic versus Microscopic Texture Characteristics . Macroscopic views (top) and corresponding microscopic views (bottom) of printed wood grain (left), concrete (middle), and terrazzo (right). The highlighted regions indicate the corresponding microscopic views, revealing distinctive surface structure even on visually uniform materials.

Fig. 3 : Gantry-based Mapping and Reference System. A microscope camera mounted to a calibrated XY gantry provides controlled motion for surface-map acquisition and independent localization queries.

Texture	Held-out RMSE ( $\mu$ m)	Held-out P95 ( $\mu$ m)
Printed Desk	25.71	54.03
Terrazzo	30.97	68.22
Concrete	26.10	55.51

TABLE I : Held-out validation of the per-texture gantry reference-frame calibration.

		L-GROUT vs. Gantry			Photometric Reference Comparison
Texture	L-GROUT Success (%)	RMSE ( $\mu$ m)	Median ( $\mu$ m)	Orient. RMSE ( ∘ )	Photo. Ref. $N$	L-GROUT–Photo. Ref. RMSE ( $\mu$ m)	Photo. Ref.–Gantry RMSE ( $\mu$ m)
Printed Desk	100.0	28.36	19.78	0.070	988	8.82	28.29
Terrazzo	100.0	57.08	50.16	0.091	997	8.86	57.69
Concrete	100.0	72.75	61.07	0.093	998	8.79	72.86

TABLE II : Global localization across three surfaces with 1,007 gantry queries per surface. L-GROUT errors are reported relative to the calibrated gantry reference. The photometric reference provides an additional image-based comparison evaluated on queries for which it returns a valid pose, with

N

denoting the number of such queries.

Fig. 4 : Robustness to Query-image Rotation. Gantry query images are synthetically rotated from

-90^{\circ}

+90^{\circ}

. (a) Translation RMSE and (b) orientation RMSE remain stable across the tested range.

Fig. 5 : Image Resolution Tradeoffs. Effect of digital downsampling on (a) localization success, (b) median translation error, (c) single-worker query latency, and (d) retained SIFT features. Precision remains stable until texture-dependent feature loss reduces localization success.

	Evaluation Coverage		L-GROUT vs. Photometric Reference
Texture	Photometric Reference Coverage (%)	L-GROUT Success (%)	Evaluation $N$	Trans. RMSE ( $\mu$ m)	Orient. RMSE ( ∘ )
Printed Desk	88.2	100.0	270	20.70	0.067
Terrazzo	99.3	100.0	304	26.42	0.101
Concrete	82.7	98.4	252	22.67	0.135

TABLE III : Mobile localization evaluated using the map-relative photometric reference. Each surface contains 306 queries across three trajectories. Errors are computed on the shared subset for which both L-GROUT and the photometric reference return a pose, with Evaluation

N

denoting the size of this subset.

Panel of Figure 9 — Fig. 6 : Thermal Expansion Measurement. Top: Experimental setup used to measure thermal expansion of the aluminum rod. Bottom-left: Representative displacement measurements from Trial 1, showing L-GROUT and direct ECC estimates compared with the thermal model, with shading indicating the heating interval. Bottom-right: Residual distributions relative to the thermal model across all three trials.

Robot Localization

Robot localization estimates a robot’s position and orientation from onboard sensors, maps, and observations of its surroundings.

See all 177 papers

A Novel Camera-to-Robot Calibration Method for Vision-Based Floor Measurements
Mar 16, 2026Jan Andre Rudolph, Dennis Haitz, Markus UlrichHand-Eye Calibration Camera Calibration
Machine Vision Metrology Lab, Institute of Photogrammetry and Remote Sensing, Karlsruhe Institute of Technology, Germany
From Language to Motion: Task-Conditioned Focal-Stack Trajectory Integration for Microscopic Robots
Oct 8, 2026Junjie Xie, Chuxuan He, Junkai Huang +6Robotics Vision-Based Robot Control
Institute of Automation, Chinese Academy of Sciences · Zhejiang Gongshang University · Safe AI Lab, Carnegie Mellon University +1
Meridian: Metric-Semantic Primitive Matching for Cross-View Geo-Localization Beyond Urban Environments
Jun 4, 2026Mason Peterson, Qingyuan Li, Yixuan Jia +4Cross-View Geo-Localization Visual Place Recognition
Massachusetts Institute of Technology, Cambridge, MA 02139, USA. · GRASP Laboratory, University of Pennsylvania, Philadelphia, PA 19104, USA. · U.S. Army Combat Capabilities Development Command, Army Research Laboratory, Adelphi, MD 20783, USA.

Mar 16, 2026J/K move · Enter open · S save

A Novel Camera-to-Robot Calibration Method for Vision-Based Floor Measurements

Jan Andre Rudolph, Dennis Haitz, Markus Ulrich

Machine Vision Metrology Lab, Institute of Photogrammetry and Remote Sensing, Karlsruhe Institute of Technology, Germany

Hand-Eye Calibration Camera Calibration

Open paper AskPDF arXiv

A novel hand-eye calibration method for ground-observing mobile robots is proposed. While cameras on mobile robots are common, they are rarely used for ground-observing measurement tasks. Laser trackers are increasingly used in robotics for precise localization. A referencing plate is designed to combine the two measurement modalities of laser-tracker 3D metrology and camera-based 2D imaging. It incorporates reflector nests for pose acquisition using a laser tracker and a camera calibration target that is observed by the robot-mounted camera. The procedure comprises estimating the plate pose, the plate-camera pose, and the robot pose, followed by computing the robot-camera transformation. Experiments indicate sub-millimeter repeatability.

Abstract

Robot Localization

Explore similar work

A Novel Camera-to-Robot Calibration Method for Vision-Based Floor Measurements

From Language to Motion: Task-Conditioned Focal-Stack Trajectory Integration for Microscopic Robots

Meridian: Metric-Semantic Primitive Matching for Cross-View Geo-Localization Beyond Urban Environments