IEEE Robotics & Automation Magazine - June 2020 - 79

feedback capabilities similar to a human physiotherapist.
These desired features are usually achieved at the expense of
other important requirements, such as transparency and
backdrivability, degrading the overall human-machine
interaction experience.
We present an active gravity compensation method that
can significantly improve the performance of mechatronic
systems used for rehabilitation and many other domains of
robotic applications. Traditional algorithms to obtain active
gravity compensation usually require the system's static equilibrium equations. However, for complex mechatronic configurations, solving these equations is not straightforward. The
use of machine learning (ML) methods can achieve gravity
compensation without the need to solve the equilibrium
equations. To validate the performance of the proposed
approach, the HomeRehab robotic rehabilitation system is
used to obtain experimental results.
Robotic Rehabilitation
Stroke is currently the second most frequent cause of death,
after coronary artery disease, and its prevalence is increasing
at an alarming rate. Hemiparesis resulting in movement disabilities is the most common outcome of stroke. Fortunately,
rehabilitation can help hemiparetic patients learn new ways of
using and moving their weak arms and legs. It is also possible
with immediate therapy that people who suffer from hemiparesis may eventually regain movement. Although there are
several approaches, extensive task-specific repetitive movement is one of the safest and most effective methods to regain
the lost mobility of the affected limbs. This therapy requires
continual medical care and intensive rehabilitation, which
often necessitates one-on-one manual interaction with the
physical therapist.
Robotic rehabilitation is an emerging field that enables a
patient to perform exercises with the assistance of a robotic
device [1]. These systems can be used to provide therapy
(even at the patient's home) for a long period of time, irrespective of skills and fatigue, compared to manual therapy.
Besides the cost-effective aspect, robotic devices introduce
higher accuracy and repeatability in performing exercises.
Precise measurements of quantitative parameters by means of
robotic instrumentation also improve the objective monitoring of patients' recovery. Furthermore, rehabilitation robots
can be combined with virtual-reality environments to engage
and push patients to keep training, as the patients' motivation
and cognitive involvement have a great impact on the outcome of rehabilitation [2].
Currently, several devices on the market provide a robotic
solution to these repetitive movements, and these have been
installed in many hospitals around the world. Some examples are InMotion ARM (Bionik Labs), ReoGo (Motorika),
and Armeo Power (Hocoma). For a successful rehabilitation
system, the robot must be able to deliver physical forces similar to manual therapy. Mechanically, this implies developing
robots with significant workspace and force feedback features. Such systems have, in turn, the drawback of being

bulky and heavy, degrading the final interaction experience
with the patient. Among the different technical challenges of
these systems, gravity compensation is key for high rehabilitation performance.
In previous work [3], we developed the HomeRehab
robotic system, which is capable of restoring haptic effects at
its handle for upper limb rehabilitation (Figure 1). This mechanism has three degrees of freedom (3 DoF) and consists of a
pantograph that pivots on a horizontal axis, but it is not perfectly gravity-balanced. Thus, an active compensation strategy
is needed. In this article, gravity compensation means removing the gravity components of the mechanical device to make
it transparent or imperceptible to the patient, in the sense
that, during the rehabilitation exercises, the patient has to
overcome only the weight of his or her own arm.
Related Work
Gravity compensation is a basic requirement in robotics, and
more specifically in haptics, to ensure system usability and
transparency. It is also a key specification of mechatronic
rehabilitation devices [4], so that users can move their arms
freely without feeling (and holding) the weight of the robot
during the therapy. In addition, friction may be a limiting factor of these mechanisms, and some strategies try to compensate for both the unknown static friction and the gravity
forces [5]. A comparison of different algorithms for gravity
compensation in parallel mechanisms can be found in [6].
In the field of rehabilitation devices, several gravity compensation strategies have been applied that are also valid for
any mechatronic system. In some cases, the mechanism is
designed to be gravity balanced, that is, to be in neutral equilibrium without requiring joint actuator torques [7]. This feature can also be achieved by adding a passive mechanism to
the robotic arm [8]. These solutions are intrinsically safe, but
they are difficult to design.
Analytical Approaches
Analytical active compensation methods balance the system,
acting on each joint according to a gravity model of the

Figure 1. The HomeRehab robotic rehabilitation system developed
at Ceit, a member of the Basque Research and Technology
Alliance (BRTA) and associated with Tecnun, Universidad de
Navarra, School of Engineering at San Sebastián.

JUNE 2020

IEEE ROBOTICS & AUTOMATION MAGAZINE

IEEE Robotics & Automation Magazine - June 2020

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2020