IEEE Robotics & Automation Magazine - March 2023 - 26

to meet the individual requirements of the end users. Through
the prism of our personal experience in the field, our aim
herein is to provide a concise description of these solutions
and assess the progress made so far. For further details on the
technical aspects, the interested reader is referred to our previous
publications in the " References " section.
INSTRUMENTED POWER WHEELCHAIRS
The hardware architecture put forward in the ADAPT project
is characterized by diversity in terms of sensing and onboard
computation (different sensor specifications, types of microcontrollers,
and so on). On the other hand, the software architecture
is unified, and it relies on Robot Operating System
(ROS) as middleware. This " lingua franca " enables the interchange
of multiple hardware and software components, which
can be tested and shared among project partners before being
integrated during the clinical trials.
In what follows, we present the wheeled platforms developed
in France and the United Kingdom and the wheelchair
simulator.
WHEELED PLATFORMS
The Aspire Create group at University College London
(UCL) has developed a smart wheeled platform by instrumenting
a Sunrise Medical Quickie Salsa M2 power wheelchair
with custom and off-the-shelf electronics (see Figure 2).
The midwheel-drive platform has six wheels with independent
suspensions, is endowed with a curb-climbing ability for
heights up to 7 cm, measures 61 cm at the widest point, and
has 60-Ah batteries that can propel it up to 10 km/h. An inerF1:
Detection
F2: Obstacle Management
F3: User Intention
F4: Shared Control
F5: Short-Term Navigation
F6: Input/Feedback to the User
F7: Long-Term Navigation
F8: Energy Management
(a)
F9: Immersiveness
F10: User Guidance (Replicate What Health
Professionals Can Do in Reality)
(b)
FIGURE 1. The functional analysis. Thanks to the collaboration with our medical partners, user needs and preferences have been
translated into a set of functional requirements. Each parent function has a number of subfunctions, some of which already have wellknown
solutions (e.g., " F1a*: positive obstacle detection " ), whereas others remain open research questions, and thus they have been
explored in greater depth within the ADAPT project (e.g., " F1b: negative obstacle detection " ). The (a) functional requirements for the
smart wheelchair and (b) additional functional requirements for the simulator.
26 IEEE ROBOTICS & AUTOMATION MAGAZINE MARCH 2023
F9a: Visual Appearance
F9b: Avatar
F9c: Motion Rendering
F9d: Ambient Sound
tial measurement unit (IMU) SparkFun 9DoF Razor, which
includes a three-axis accelerometer, gyroscope, and magnetometer,
has been installed under the driver's seat. Industrial
wheel encoders (Kubler 500 ppr) together with 3D-printed
pulleys have been placed in the narrow space between the
main drive wheels and the chassis of the wheelchair to obtain
measurements from odometry. Twelve ultrasonic sensors
(SRF08) have been installed in four custom housings in the
corners of the chassis of the wheelchair. Each housing contains
three ultrasonic sensors covering a theoretical angle of
135º, where obstacles can be detected. Electric current sensors
and voltage measurements are used to monitor the electric
power flow through the two motors. Finally, a
single-board computer (Raspberry Pi 3B+) acts as the ROS
master, using a publisher-subscriber model. We refer the
reader to [16] for more details on all these components,
including the schematics of the hardware architecture.
The wheeled platform developed at the Institut National des
Sciences Appliquées (INSA) Rennes also builds on the Quickie
Salsa M2 wheelchair. It is equipped with 48 time-of-flight
(ToF) sensors organized in seven modules distributed along
its perimeter: six modules of six sensors are located on each
side and under the footplates, and one module of 12 sensors is
installed behind the backrest (see Figure 3). The ST VL53L1X
sensors have the following technical specifications: distance
measurement up to 4 m, ranging frequency up to 50 Hz, typical
full field of view of 27º, and size of 4.9 × 2.5 × 1.56 mm3.
Their measurements have been used to directly detect positive
obstacles (doors, walls, and so on) around the wheelchair or
to infer the presence of negative obstacles (potholes, inclines,
F1a*: Positive Obstacle
F1b: Negative Obstacle
F1c: 3D Obstacles
F1d: Mobile Obstacle
F1e: Key Frames
F1f: Door
F1g: Curb Ramp
F1h: Specific Object

IEEE Robotics & Automation Magazine - March 2023

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