IEEE Robotics & Automation Magazine - September 2018 - 96

decisions during the sample return mission. This relationship
can be broken into multiple perception problems, such as
localization of the robot body, mapping of the environment,
detection of nearby obstacles or hazards, and identification
and localization of samples.
The navigation/homing subsystem was designed to give
Cataglyphis the basic capability to reach ROIs accurately and
return home without getting lost. This was achieved mainly
through dead reckoning using IMUs, wheel encoder measurements [2], and a homing beacon for navigation reference
whenever possible. A simultaneous localization and mapping
(SLAM) algorithm provided a backup localization capability in
addition to generating maps for path planning and supporting
human decision making during a communication update.
The design of the homing beacon occurred concurrently
with the robot design, as shown in Figure 3, with two main
considerations: 1) can Cataglyphis reliably detect the homing
beacon at a sufficient range, and 2) can Cataglyphis accurately
estimate its pose using the homing beacon? The homing beacon detection range needed to be greater than the worst-case
dead-reckoning error during the challenge.
The accuracy requirement of the homing beacon update
was motivated by the following factors. First, if Cataglyphis
was able to detect the homing beacon, it must be able to return

Mapped SLAM Features

Prior Map

Human
Input

GUI

home with high certainty using successive homing updates.
Therefore, it was not essential that the pose estimate be accurate at far distances, but the direction to home was critical. At
close ranges, the pose accuracy became highly important,
because Cataglyphis had to drop samples off on the platform
in precise locations to prevent intersample contact. Finally,
because the initial robot heading angle error would compound
into large dead-reckoning errors, an accurate robot pose estimate was needed each time Cataglyphis was near the starting
platform to accurately reach ROIs at the far ends of the field.
A passive homing beacon design, consisting of two vertical
cylinders, was used in 2016, which could be identified by lidar
according to their shapes and reflectivity. With this design, Cataglyphis could detect the homing beacon as far away as 30 m,
with an increased probability of detection and increased accuracy of the pose estimate when it was closer to the homing beacon.
To enhance the robustness of homing beacon detection, the
expected growth of the heading error during dead reckoning
was tracked by Cataglyphis as a reference. A homing beacon
update would be postponed if the difference between the heading angle estimate from dead reckoning and the heading angle
derived from a homing beacon observation was greater than 5°.
This may indicate either a false homing beacon detection or a
large drift in the robot dead-reckoning solution.

Map
Manager

Known
Hazard
Locations

Safe Pathing

Paths Around
Hazards

ROI Locations

Emergency
Conditions

Mission
Planning

Human Input
Robot Pose

SLAM

Actions to Perform

Obstacle Avoidance Maneuvers
Processed Point
Cloud Features
Lidar
Processing

HSM

Dead-Reckoning
Pose Estimate
Homing
Beacon
Scans

Robot Pose

Navigation/
Homing
IMU and
Wheel Encoder
Measurements

Laser Point
Clouds

Computer
Vision

Images

Detected
Sample
Information

Executive

Commands to Actuators

Hardware Interfaces

Figure 5. The Cataglyphis software architecture. Information received through the hardware interfaces and the graphical user interface
(GUI) was processed by various perception nodes to support decisions made by the mission-planning node. It then outputted actions
for the robot to perform, which was handled by the executive node. A health and status management (HSM) node helped with
recovering from faults [2].

IEEE ROBOTICS & AUTOMATION MAGAZINE

september 2018

IEEE Robotics & Automation Magazine - September 2018

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2018