IEEE Robotics & Automation Magazine - March 2021 - 74

0.54 m. The robot is equipped with a frontal laser rangefinder and three rear sonars that are used to scan the environment. The rangefinder has a π-rad field of view and an
angular step of 0.006 rad, with a visible range of 0.05-10 m,
operating at a rate of 15 Hz. The robot autonomously navigates its environment, thanks to the embedded sensors. To
address the two intended scenarios of logistics and disinfection with the same platform, a modular system was
developed that could be adapted as needed:
● For the logistics scenario, the UCBM COVID robotic
system was equipped with a box behind its head [Figure 2(c)] and was able to deliver materials weighing up
to 5 kg.
● For disinfecting the hospital environment, it was
equipped with a UV-C lamp [Figure 2(d)].
Fundamental TIAGo capabilities, e.g., chassis motion
and autonomous navigation, were exploited to promptly
address the pandemic through available technology. On
the other hand, TIAGo is a sophisticated robotic system
capable of interacting with human users in multiple
modalities and performing reaching and grasping tasks.
This opens opportunities for future developments in hospital scenarios (which we are working on), where all the
robot's functionalities can be fully exploited.
Two ad hoc, easy-to-use web graphical user interfaces
(GUIs) were developed in HTML to select commands for
each of the scenarios (the GUIs are described in the " Scenario 1: Logistics " and " Scenario 2: Disinfection " sections). They ran on a laptop mounted to the back of
TIAGo [Figure 2(b)]. The communication between the
robot and the computer was established via an Ethernet
port to avoid data packet losses and data slowdowns from
wireless communication. The GUIs could be easily
launched from the computer through any browser and
operating system. Once a command was selected by a
user, the robot autonomously accomplished the desired
task. The energy autonomy of the UCBM COVID platform runs out in 4 h. Hence, the system needs to be
recharged multiple times per day. The time required for a
full recharge is approximately 2 h. The platform was
equipped with a 23.4-Ah portable generator providing
power (220 V) to the UV-C lamp. The external power
supply is shown in Figure 2(d).
Localization, Mapping, and
Path Planning
To perform autonomous navigation, the mobile robot is
equipped with a simultaneous localization and mapping
(SLAM) algorithm. It requires data from the exteroceptive
sensors (i.e., lidars, sonars, and cameras) as well as odometry information provided by wheel encoders and inertial
measurement units. The SLAM algorithms estimate the
robot pose xt inside known and unknown environments
and build a map mt. This is performed by collecting and
fusing sensor measurements zt and odometry readings ut,
with all these quantities indexed by a time step t.
74

IEEE ROBOTICS & AUTOMATION MAGAZINE

MARCH 2021

The Gmapping [15] strategy was implemented to solve
the SLAM problem. Such a planar technique relies on
robot odometry as well as measurements coming from
exteroceptive sensors (e.g., sonars and lasers) to estimate
the robot's pose and the map where the robot is operating.
The map is represented as a planar grid (the occupancy
grid) consisting of a set of square cells, each having a probability value of the navigability of the cell itself. A Bayesian
approach enables fusing pose transitions and laser scans in
a probabilistic way. The algorithm solves the SLAM by factorizing the posterior probability with a Rao-Blackwellized particle filter (RBPF) [15]. Such a probability [16]
is expressed as
p (x t, m | z t, u t) = p (x t | z t, u t) p (m t | x t, z t),

(1)

where a complexity reduction has been performed through
the RBPF factorization. Here, p (x t | z t, u t) is about having
pose xt, given measurements zt and odometry ut, i.e., a
pure localization problem; p (m t | x t, z t) represents the
map mt, given robot pose xt and scan measurements zt, i.e.,
a pure mapping problem.
The robot's pose xt and the map of the environment mt
are stored inside a set of particles through a particle filter
approach. The ith particle has the history of previous robot
poses xi, the map computed from such a history mi, and a
weight wi. The weight is the probability that the particle
represents both the pose and the map. Gmapping estimates
this probability based on robot odometry and scan matching between consecutive laser scans and generates new particles. At the same time, the number of particles N is kept
constant to reduce the computational burden. The complexity of the adopted approach is O (N # M), where M is
the size of the generated grid map, as described in [15].
To address autonomous navigation, the robot is able to
generate a reference trajectory through a path-planning
module. This module is based on a Robot Operating System navigation package called global_planner. It needs the
map to preplan the path to be followed and a local planner based on the global dynamic window approach,
described in [17], that recalculates the trajectory according to the instantaneous local condition along the path.
This is an obstacle-avoidance approach that evaluates the
robot's kinematic and dynamic constraints and facilitates
identifying the desired velocities. The desired velocities
are chosen to maximize the alignment of the robot with
the target and minimize the length of the trajectory in the
absence of any obstacle. In this way, it is possible to combine path planning and obstacle avoidance to enable the
robot to navigate safely in an unstructured environment.
The navigation parameters adopted in this work are given
in Table 1. They are used to plan the global path, adjust
the local trajectory, and adapt the behavior of the platform
in the presence of obstacles.
The Gmapping SLAM algorithm was chosen because of
two features, i.e., the ease of implementation and

IEEE Robotics & Automation Magazine - March 2021

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - March 2021