IEEE Robotics & Automation Magazine - December 2023 - 95

significant performance improvement.
The FI was 0.04 m. As a result, whenever
the inflated footprint overlaps any
detected obstacles, the robot transitions
into recovery behaviors. Two illustrations
of the inflated footprint are
depicted in Figure 4(a), in which the
green region and red region indicate
safe and unsafe conditions, respectively.
REAL-WORLD APPROACH
The only difference between simulation
and real-world methods was in the safety
check when moving forward. An
MPC approach was adopted for the
safety check during forward movement
in the real world. MPC not only takes
into account the robot's current footprint
but also integrates future position
information, enabling the robot to proactively
plan its movement and adjust
its trajectory accordingly. Figure 4(b)
depicts this process, with green indicating
safe portions of the trajectory and
red showing a detected future collision.
UNIVERSITY OF ALMERIA
The University of Almeria team's implementation
was built upon the Mobile
Robot Programming Toolkit (MRPT),
an open source C++ framework specifically
developed for robotics applications,
including libraries for navigation. The
MRPT ROS nodes were fine-tuned to
align with the specifications of the
Clearpath Robotics Jackal robot. A notable
difference between the simulation
qualifier and physical finals is that in
simulation the goal was well known,
with fixed coordinates in a map known
a priori. By contrast, in the physical
finals there were no absolute coordinates
of the goal. In practice, this makes the
real-world navigation close to pure
exploration. Those caveats aside, the
architecture of the system comprises the
subsystems enumerated in the following
paragraphs, each one implemented as an
independent ROS node.
LOCALIZATION
A custom implementation of particle
filter-based localization with an adaptive
number of particles using the Kullback-Leibler
distance approach was
used by the University of Almeria team
and is capable of using several metric
maps at once for localization.
LOCAL OBSTACLES MAP
The purpose of this node is to perform
real-time acquisition and processing of
sensor data, in this case the lidar scans.
These raw sensor data were subsequently
transformed into a 2D point
cloud representation, localized within
the robot's own coordinate frame and
decimated into lower resolution for
faster processing.
PATH PLANNER (HIGHER LAYER)
At a relatively low rate (1 Hz or slower),
the local obstacles were considered
to find a kinematically feasible path
using a path planning algorithm, which
was then sent to a local path follower.
The team's path planner used a custom
algorithm based on A* on a discrete
lattice of the state-vector space of the
vehicle, i.e., the SE(2) pose plus velocities.
Arcs between the lattice nodes
were explored efficiently using parameterized
trajectory generators (PTGs),
a concept derived from past works,
which defines families of paths to help
explore the environment with kinematically
and dynamically feasible paths.
LOCAL PLANNER (LOWER LAYER)
Once a path is found, the task of generating
motor commands to follow it,
including avoiding any new obstacles, is
accomplished by a reactive navigation
system, which is also based on the trajectory
parameter space (TP-Space). In
TP-Space, the robot, regardless of its
physical shape and kinematic constraints,
is transformed into a free-flying
point within a newly formulated parameter
space. This transformation incorporates
the robot's shape and kinematic
restrictions, thereby allowing for efficient
navigation by taking into account
the robot's specific physical characteristics
and movement capabilities. On the
other hand, PTGs define a set of potential
trajectories for the robot, parameterized
by variables such as path shape,
Front
270°
Left
Right
(a)
Rear
(b)
FIGURE 3. (INVENTEC) (a) Costmap with robot rear ROI (yellow rectangle). (b) Jackal
lidar field of view.
Goal
(a)
(b)
FIGURE 4. (INVENTEC) (a) FI for collision detection. (b) MPC footprint forward safety check.
DECEMBER 2023 IEEE ROBOTICS & AUTOMATION MAGAZINE
95
Rear

IEEE Robotics & Automation Magazine - December 2023

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2023