IEEE Robotics & Automation Magazine - March 2023 - 29

classical data processing pipeline, Metashape relies on GNSS
(Global Navigation Satellite System) measurements, which are
typically available outdoors but not indoors. To address this
issue in indoor environments, we first generated a sparse 3D
model with the associated camera poses using OpenVSLAM
[20], an off-the-shelf software package for visual simultaneous
localization and mapping. These poses are then given as input
to Metashape. The trajectory of the wheelchair estimated with
OpenVSLAM and an external motion capture system (see the
" Driving Assistance " section) were used to assess the quality
of the driving assistance provided by SpheriCol [21].
OpenVSLAM may fail if the interframe motion between
successive images in a sequence is large, which might result,
for example, from an abrupt change in the joystick position.
Hence, it cannot be directly used online to assist wheelchair
users. In [22], we overcame this limitation by proposing a new,
accurate direct visual gyroscope, which copes with large interframe
motions. Based on the mixture of photometric potentials,
it takes the spherical images from a twin-fisheye camera
as input and provides an estimate of its 3D orientation with
respect to a reference image (typically the initial one). In our
experiments, we observed no performance degradation for reference
images captured up to several tens of meters away and
for rotational displacements of a few tens of degrees. To quantitatively
evaluate the performance of our visual gyroscope
and other state-of-the-art vision-based ego-motion estimation
algorithms, a data set of omnidirectional images captured by
catadioptric and twin-fisheye cameras mounted on different
robotic platforms, PanoraMIS [23], has been made publicly
available. In particular, Sequence 7 of the data set (1.35 km,
in whole) comes with an accurate ground truth provided by
an Adept MobileRobots Seekur Jr. robot (with integrated and
external IMUs, GPS measurements, and wheel odometry).
This robot was chosen since its footprint is comparable to that
of a standard power wheelchair.
SHARED CONTROL
Shared control is a concept involving collaboration between a
human and a machine [2], [3]. The human expresses an intention
that the machine facilitates and implements in an optimal
way. The assist-as-needed paradigm provides assistance
only when required, providing the user with as much control
authority as possible. This concept is a key emerging technology
with wide applicability in medical robotics.
Two modular and complementary shared control strategies
have been proposed in the ADAPT project to cater for the
wide variety of user needs: the first strategy is sensor based
and the second is model based, as illustrated in Figure 7. Both
strategies are compatible with new vibration-based human-
machine interfaces, as detailed in the " Human-Machine
Interaction and Haptic Feedback " section.
SENSOR-BASED SHARED CONTROL
The sensor-based shared control method developed at INSA
Rennes relies on a simple geometric algorithm that can be
easily implemented on low-cost embedded devices with limited
computational resources [24]. The algorithm makes use
of measurements from any type of range sensor on the wheelchair
and leverages the distance constraints to compute two
areas in the velocity space, which correspond to the input
velocities of the wheelchair that are safe (allowed area) or
unsafe (forbidden area). The shared control blends the user's
input and algorithm's output to ensure safe and smooth
wheelchair navigation. This obstacle avoidance solution is
Environment
Distance to
Obstacles
Shared Control
Sensor Based
Model Based
(υjoy, ωjoy)
1
Driver Input
Joystick Data
0.5
-0.5
-1
10 -1
ωjoy
Wheelchair
FIGURE 7. The shared control working principle. Using information from the environment and driver, restrictions (red-shaded area) are
created in the joystick plane, yielding safe linear and angular velocities (, )
y~ for the wheelchair.
joyjoy
MARCH 2023 IEEE ROBOTICS & AUTOMATION MAGAZINE
29
υjoy

IEEE Robotics & Automation Magazine - March 2023

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