OPENTECHNICAL PUBLICATION
OT-2026-000000016

Two-Stage Fiducial Docking for Autonomous Mobile Robots

Published: 2026-09-15T19:01:17.067147Z
Permanent URL: https://opentechnical.org/p/OT-2026-000000016
Categories: Robotics & Automation > Autonomous Mobile Robots > Docking and Charging

------------------------------------------------------------------------

ABSTRACT

This disclosure describes a system for autonomous mobile robot docking
utilizing a two-stage fiducial approach. It addresses the common challenge
where a single fiducial either provides a wide capture range or high
precision, but not both, often leading to docking failures during the
transition. The system employs a larger, easily detectable coarse fiducial for
initial approach and target acquisition, followed by a smaller, high-precision
fine fiducial for final alignment and accurate docking. This method enables a
broad capture envelope while ensuring precise final positioning, facilitating
reliable and robust autonomous docking operations.

1. Technical Field

[0001] The technical field relates to autonomous mobile robots (AMRs) and
their ability to autonomously navigate to and precisely connect with a
designated docking station or charging point. This includes methods for
localization, pose estimation, and control specifically adapted for
close-range interaction and physical coupling.

2. Technical Context

[0002] Autonomous mobile robots require reliable and precise docking
capabilities for tasks such as battery charging, material transfer, or data
exchange. A common challenge in achieving autonomous docking is the trade-off
between the capture range and the precision offered by visual fiducials. A
single fiducial designed for a wide capture range, meaning it can be detected
from a long distance or wide angle, typically lacks the geometric features
necessary for high-precision pose estimation at close range. Conversely, a
fiducial optimized for high precision often has fine features that are
difficult to detect or resolve from a distance, leading to a small capture
range. When using a single fiducial, the transition from a wide approach to a
precise final alignment is prone to failure, as the robot may lose track of
the fiducial or be unable to refine its pose sufficiently for docking. This
limitation can result in unsuccessful docking attempts, increased operational
time, and potential damage to the robot or the docking station.

3. System Overview

[0003] The described system employs a two-stage fiducial approach to overcome
the limitations of single-fiducial docking. An autonomous mobile robot (10) is
equipped with an onboard camera (14) and a robot controller (20). The docking
station (12) presents two distinct fiducial markers: a coarse fiducial (16)
and a fine fiducial (18). The coarse fiducial (16) is designed for detection
over a wide range of distances and angles, enabling the robot (10) to navigate
into a general docking area. Once the robot (10) is within a defined proximity
to the docking station (12), it transitions to detecting and tracking the fine
fiducial (18). The fine fiducial (18) provides highly accurate pose
information for precise alignment and final docking maneuvers. The robot
controller (20) manages the handover between these two stages, utilizing
algorithms for fiducial detection, pose estimation, and motion control. FIG. 1
illustrates the overall system components and their interactions, while FIG. 3
depicts the operational control flow.

4. Components

[0004] The system comprises several key components:

[0005] The Autonomous Mobile Robot (AMR) (10) is a wheeled or tracked platform
capable of autonomous navigation and executing motion commands. It carries the
necessary sensors and computational resources for docking.

[0006] The Docking Station (12) is a fixed or semi-fixed structure designed to
receive the AMR (10). It physically incorporates the coarse fiducial (16) and
the fine fiducial (18) in a known relative arrangement. The docking station
(12) may also incorporate mechanical and electrical interfaces for coupling
with the robot (10).

[0007] The Onboard Camera (14) is a digital camera mounted on the AMR (10),
typically forward-facing. It captures images of the environment, including the
fiducial markers, and transmits them to the robot controller (20). The camera
(14) is specified to have a resolution sufficient for distinguishing the
features of both fiducials within their respective detection ranges. A typical
field of view might range from 60 to 90 degrees horizontally.

[0008] The Coarse Fiducial (16) is a visual marker affixed to the docking
station (12). It is characterized by large, distinct patterns or high
contrast, making it easily detectable from a significant distance and over a
wide angular range by the onboard camera (14). Examples include large AprilTag
variants, QR codes, or checkerboard patterns. Its primary function is to guide
the robot (10) to a pre-defined approach zone, providing approximate pose
information. As shown in FIG. 2, the coarse fiducial (16) is typically much
larger than the fine fiducial (18).

[0009] The Fine Fiducial (18) is a visual marker, also affixed to the docking
station (12), positioned in a known relation to the coarse fiducial (16). It
features smaller, more intricate patterns or higher feature density, allowing
for highly precise pose estimation when viewed at close range. Examples
include concentric circles, specific dot patterns, or smaller, high-resolution
AR markers. Its function is to provide sub-millimeter level accuracy for the
final docking maneuver. As shown in FIG. 2, the fine fiducial (18) is
typically centrally located or offset by a specific distance (D) relative to
the coarse fiducial (16).

[0010] The Robot Controller (20) is an embedded computing system on the AMR
(10). It processes image data from the onboard camera (14), executes fiducial
detection and pose estimation algorithms, manages the state transition between
coarse and fine tracking, and generates motion commands for the actuator
system (24). It typically incorporates a proportional-integral-derivative
(PID) control loop or similar feedback mechanism for tracking and alignment.

[0011] The Localization Sensor (22) (e.g., LiDAR, odometry, inertial
measurement unit) provides general robot (10) pose information. This sensor
(22) aids in navigating to the general vicinity of the docking station (12)
and acts as a fallback for pose estimation if fiducial tracking is temporarily
lost.

[0012] The Actuator System (24) comprises the drive motors, steering
mechanisms, and associated motor controllers that enable the AMR (10) to
execute translational and rotational movements based on commands from the
robot controller (20).

5. Operation

[0013] The docking operation proceeds through distinct stages as outlined in
FIG. 3.

[0014] Initial Approach (100): The AMR (10) begins its journey to the general
docking area using its primary navigation system, which may involve global
localization sensors (22) such as LiDAR or GPS, combined with odometry. The
robot (10) moves towards a known approximate location of the docking station
(12).

[0015] Coarse Target Search (102): As the robot (10) approaches the designated
area, the onboard camera (14) continuously captures images. The robot
controller (20) initiates a search for the coarse fiducial (16) within the
camera's field of view. The coarse fiducial's (16) large size and distinct
pattern facilitate its detection even when the robot (10) is several meters
away and at a significant angular offset from the docking station (12).

[0016] Coarse Target Tracking (104): Once the coarse fiducial (16) is
detected, the robot controller (20) calculates the robot's (10) relative pose
(position and orientation) to the docking station (12) using the coarse
fiducial (16) data. This pose information is used to generate motion commands
for the actuator system (24) to guide the robot (10) towards a closer, more
centered approach to the docking station (12). The robot (10) maintains
tracking of the coarse fiducial (16) and continuously refines its pose
estimate.

[0017] Transition Threshold (106): As the robot (10) closes in on the docking
station (12), its estimated distance to the target or the apparent size of the
coarse fiducial (16) on the image plane reaches a pre-defined threshold. At
this point, the robot controller (20) prepares for the transition to fine
tracking. This threshold is typically set such that the fine fiducial (18) is
guaranteed to be within the camera's (14) field of view and resolvable for
accurate detection. A typical transition distance might be 0.5 to 1.5 meters
from the docking point.

[0018] Fine Target Search (108): Upon reaching the transition threshold (106),
the robot controller (20) initiates a search for the fine fiducial (18). Since
the relative position of the fine fiducial (18) to the coarse fiducial (16) is
known (offset D, as seen in FIG. 2), the search area for the fine fiducial
(18) can be constrained, reducing computational load and improving search
reliability.

[0019] Fine Target Tracking (110): Once the fine fiducial (18) is detected,
the robot controller (20) switches its primary pose estimation and control
loop to use the data from the fine fiducial (18). This stage provides highly
precise pose updates, allowing for sub-millimeter positioning accuracy. The
robot (10) continues to adjust its position and orientation using the actuator
system (24) to align precisely with the docking station (12).

[0020] Final Docking (112): With precise alignment achieved through fine
target tracking (110), the robot (10) performs the final docking maneuver.
This may involve moving forward a short, fixed distance to engage mechanical
or electrical connectors, such as charging pins or mechanical latches.

[0021] Recovery from Lost Target Mid-Approach: If the coarse fiducial (16) is
lost during coarse tracking, the robot controller (20) can use its last known
pose relative to the docking station (12) and data from the localization
sensors (22) to attempt to re-acquire the coarse fiducial (16) by executing a
pre-defined search pattern (e.g., small lateral sweeps). If the fine fiducial
(18) is lost during fine tracking, the system can revert to coarse tracking if
the coarse fiducial (16) is still visible, or fall back to a more aggressive
re-acquisition strategy similar to the coarse stage. If both are lost or no
recovery is possible, the robot (10) may initiate a full re-approach from a
safe distance or alert an operator.

[0022] Behavior when the Dock is Partially Occluded: If the coarse fiducial
(16) is partially occluded, its design (e.g., large size, redundant patterns)
allows for robust detection even with a significant portion obscured. The pose
estimation algorithm can still derive a usable, albeit less precise, pose. If
the fine fiducial (18) is partially occluded during its tracking stage, the
robot (10) might attempt to shift its position slightly to clear the
occlusion, or if the occlusion is severe enough to prevent reliable pose
estimation, it will revert to coarse tracking or initiate a recovery
procedure. The system prioritizes maintaining some level of visual tracking
over proceeding blindly.

6. Example Embodiments

[0023] Embodiment 1: This embodiment utilizes an approximately 500 mm by 500
mm retroreflective checkerboard pattern as the coarse fiducial (16). The
pattern is constructed from alternating squares of retroreflective and
non-reflective material. The onboard camera (14) is equipped with an
integrated LED illuminator to enhance visibility of the retroreflective
elements. The fine fiducial (18) is a 100 mm diameter target featuring three
concentric circles, where the outer two circles are black and the innermost
circle is white. The robot controller (20) uses a blob detection algorithm for
the coarse fiducial (16) to identify large reflective areas and estimate
approximate centroid and orientation. For the fine fiducial (18), a Hough
transform is applied to detect the circular patterns, enabling precise
sub-pixel center detection and calculation of the relative offset.

[0024] Embodiment 2: In this embodiment, the coarse fiducial (16) is an
AprilTag (e.g., Tag36h11) with a side length of 400 mm, printed on a
high-contrast background. The fine fiducial (18) is an active LED array
arranged in a distinct pattern (e.g., a cross with four corner points and a
central point) covering a 150 mm square area. The onboard camera (14) is a
global shutter camera, and the robot controller (20) is configured to detect
the AprilTag using a standard library for coarse tracking. For fine tracking,
the controller (20) employs image processing to identify the centroids of the
illuminated LEDs, calculating the geometric transformation based on the known
physical arrangement of the LED emitters. This active illumination can improve
robustness in varying lighting conditions.

[0025] Embodiment 3: This embodiment integrates a depth camera as the onboard
camera (14). The coarse fiducial (16) is a large, geometrically distinct shape
(e.g., a pyramid or a large "H" shape) with a base dimension of approximately
600 mm, providing clear depth contours. The fine fiducial (18) is a smaller,
high-resolution grayscale pattern (e.g., a dot grid with 5 mm pitch) printed
on a flat surface. For coarse tracking, the depth camera (14) extracts 3D
point cloud data from the coarse fiducial (16) and performs 3D object
registration to estimate the robot's (10) pose. For fine tracking, 2D image
processing techniques are applied to the grayscale pattern of the fine
fiducial (18) from the depth camera's (14) RGB stream, leveraging the depth
information to validate the pattern's planarity and refine the pose estimate
for improved accuracy.

7. Alternative Configurations

[0026] The choice of fiducial types can be varied. Instead of passive visual
markers, the coarse fiducial (16) could be an array of ultrasonic emitters or
a set of inductive coils, with corresponding receivers on the robot (10) for
long-range detection. The fine fiducial (18) could be a laser projection
system emitting a precise pattern onto the robot's (10) front face, which is
then analyzed by a camera (14) on the robot (10). Alternatively, the fine
fiducial (18) could be an active light source array, blinking in a specific
sequence to encode additional identification information or improve detection
in challenging lighting. The onboard camera (14) could be a stereoscopic
vision system, providing its own depth information and potentially negating
the need for a separate localization sensor (22) for close-range pose
estimation. The docking station (12) itself might not be fixed; in some
configurations, it could be a mobile platform that aligns itself to a
stationary robot (10) using a similar two-stage approach, reversing the roles
of mobile and stationary components.

8. Parameter Variations

[0027]
  Parameter | Range | Effect of Moving Towards Lower End | Effect of Moving Towards Upper End
  Coarse Fiducial Size | 200 mm to 1000 mm | Reduced maximum detection distance, smaller initial capture range. | Increased maximum detection distance, larger initial capture range.
  Fine Fiducial Size | 50 mm to 200 mm | Improved precision at very close range, reduced detection distance for fine tracking. | Reduced potential for very fine precision, increased detection distance for fine tracking.
  Camera Field of View (FoV) | 45° to 120° | Narrower view requires more accurate initial robot positioning. | Wider view increases initial target acquisition range, introduces lens distortion.
  Approach Speed | 0.1 m/s to 0.8 m/s | Increased time to dock, potentially higher precision during final approach. | Reduced time to dock, higher risk of overshoot or reduced precision.
  Transition Distance | 0.3 m to 2.0 m | Earlier switch to fine tracking, requiring more robust fine fiducial detection at range. | Later switch to fine tracking, requiring more robust coarse fiducial tracking.
  Fiducial Contrast Ratio | 5:1 to 20:1 | Reduced reliability in varied lighting, higher chance of detection failure. | Improved reliability in varied lighting, better performance with camera auto-exposure.
  Robot Controller Cycle Time | 10 ms to 100 ms | More responsive control, better tracking of dynamic targets. | Slower response, potential for oscillation or loss of track.

9. Additional Implementations

[0028] This two-stage fiducial docking approach can be extended to scenarios
beyond simple charging. It can facilitate automated tool changing for robotic
arms mounted on mobile platforms, where precise alignment of the robot base is
critical before tool engagement. The system can be adapted for autonomous
loading and unloading of materials, where the robot (10) needs to align with a
conveyor belt or storage rack. For environments with multiple docking stations
(12), each station could incorporate a unique coarse fiducial (16) identifier
(e.g., different AprilTag families or colors) to direct the robot (10) to a
specific dock, followed by a generic fine fiducial (18) for precise alignment.
The fiducials could also be integrated into dynamic targets, such as moving
platforms or other robots, requiring the system to track and dock with a
target that is itself in motion. Furthermore, the principles can be applied to
re-positioning modules or payloads within a larger system, where one component
acts as the 'robot' and another as the 'docking station'.

10. Figures

FIG. 1
    A block diagram illustrating the primary components and information flow
    in the two-stage docking system.

FIG. 2
    An elevation view showing the relative positioning and geometry of the
    coarse and fine fiducials on the docking station.

FIG. 3
    A simplified control flow diagram illustrating the sequence of operations
    during the two-stage docking process.

FIGURES

FIG. 1  A block diagram illustrating the primary components and information flow in the two-stage docking system.
FIG. 2  An elevation view showing the relative positioning and geometry of the coarse and fine fiducials on the docking station.
FIG. 3  A simplified control flow diagram illustrating the sequence of operations during the two-stage docking process.

------------------------------------------------------------------------

OpenTechnical publication OT-2026-000000016
Published 2026-09-15T19:01:17.067147Z
https://opentechnical.org/p/OT-2026-000000016