The Three Laws of Neurorobotics: A Review on What Neurorehabilitation Robots Should Do for Patients and Clinicians
J. Med. Biol. Eng.
The Three Laws of Neurorobotics: A Review on What Neurorehabilitation Robots Should Do for Patients and Clinicians
Marco Iosa 0 1 2
Giovanni Morone 0 1 2
Andrea Cherubini 0 1 2
Stefano Paolucci 0 1 2
0 CNRS-UM LIRMM UMR 5506, IDH group , 161 Rue Ada, 34392 Montpellier , France
1 Clinical Laboratory of Experimental Neurorehabilitation, IRCCS Fondazione Santa Lucia , Via Ardeatina 306, 00179 Rome , Italy
2 & Marco Iosa
Most studies and reviews on robots for neurorehabilitation focus on their effectiveness. These studies often report inconsistent results. This and many other reasons limit the credit given to these robots by therapists and patients. Further, neurorehabilitation is often still based on therapists' expertise, with competition among different schools of thought, generating substantial uncertainty about what exactly a neurorehabilitation robot should do. Little attention has been given to ethics. This review adopts a new approach, inspired by Asimov's three laws of robotics and based on the most recent studies in neurorobotics, for proposing new guidelines for designing and using robots for neurorehabilitation. We propose three laws of neurorobotics based on the ethical need for safe and effective robots, the redefinition of their role as therapist helpers, and the need for clear and transparent human-machine interfaces. These laws may allow engineers and clinicians to work closely together on a new generation of neurorobots.
Rehabilitation; Robotic training; Neuroscience; Ethics; Medical robots
1.1 Controversial Effectiveness of Robots for Neurorehabilitation
The first robots used for neurorehabilitation were
developed in the 1980s [
], their potential was claimed in the
], and robotic exoskeletons started to spread in
the 2000s [
]. However, their is still debate on the
effectiveness of robots in neurorehabilitation.
Contrasting results were obtained in different studies
about neurorehabilitation robot efficacy [
though the results of some randomized controlled trials
performed on wide samples showed significant
improvements in the outcome of robot-assisted therapy with respect
to usual care [
]. Meta-analyses have only partially
helped in clarifying the objective effectiveness of robotic
training, with most results being inconclusive. A 2008
Cochrane review on post-stroke arm training robots [
concluded its analysis on 11 studies (328 subjects) by
stating that: ‘‘patients who receive electromechanical and
robot-assisted arm training after stroke are not more likely
to improve their activities of daily living, but arm motor
function and strength of the paretic arm may improve’’.
The same authors further updated their Cochrane review in
], including 19 trials (666 subjects), concluding:
‘‘Patients who receive electromechanical and robot-assisted
arm training after stroke are more likely to improve their
generic activities of daily living. Paretic arm function may
also improve, but not arm muscle strength’’. These results
were hence in opposition with those obtained previously.
Although the second Cochrane review should be
considered more reliable, given the higher number of trials and
enrolled subjects, the contrasting results (also in terms of
muscle strength) lead to confusion.
Cochrane reviews on walking rehabilitation performed
using robots also provide inconsistent results. A Cochrane
review, as well as its update [
], reported higher
probability of recovery in patients who receive
electromechanical-assisted gait training in combination with
physiotherapy, whereas another Cochrane review 
reported similar recovery probabilities for patients with and
without treadmill training (i.e., with and without body
Besides effectiveness, three other aspects deserve
attention. Firstly, these Cochrane reviews analysed
electromechanical devices and robots as a single and
homogeneous field. In fact, electromechanical devices developed
for neurorehabilitation (e.g., treadmill with body weight
support or Gait Trainer (Reha-Stim, Berlin, Germany)) are
often but improperly considered members of the robot
]. This is a major concern for the designers of
robot-therapy systems, who have failed so far to provide a
comprehensive and agreed-on framework for the correct
classification of these devices [
]. A second aspect
deserving attention is that many studies about the efficacy
of specific devices were published after their
commercialization. This approach is inconceivable in other medical
fields, for example pharmacology. The third point to take
into account is that effectiveness should be referred not
only to the device per se, but also to the specific patient
groups targeted by the therapy [
], and to the timing
and protocol adopted for that device [
]. This point was
highlighted by Mehrholz et al. [
]: the correct use of new
technologies must rely on the information regarding the
types of patients and the phase of rehabilitation that will
benefit from specific technologies. For example, patients
with more severe impairments in the motor leg can benefit
more from robotic-assisted therapy, in combination with
conventional therapy, than from conventional therapy
alone. This likely occurs because, in the case of very
impaired patients, robotic devices, increase the therapy
intensity with respect to conventional ones [
Conversely, patients with greater voluntary motor function in
the affected limb can perform intensive training also in
conventional therapy. For these patients,
neurorehabilitators may prefer less constrained, more ecological, and more
variable exercises . Physical condition is not the only
factor determining the best class of neurorobot users: the
patient psychological profile can also be important in
attaining superior motor outcomes with robot training
compared to conventional therapy [
These results have led to a proposal of a change in the
research question about the effectiveness of robot devices:
‘‘instead of asking ourselves whether robotic devices are
effective in rehabilitation, we should determine who will
benefit more from robotic rehabilitation’’ [
and exclusion criteria are not the only characteristics to be
determined in the design of a rehabilitation protocol when a
robot is used. Few studies have focused on the definition of
guidelines for an effective selection of movement
parameter values (such as joint angles, speeds, applied forces, and
torques) and for better timing of robot therapy
administration, both tailored on the patient’s capacities and needs.
However, before further discussing the issue of
effectiveness, and the reasons of the limited credit that is given
to neurorobots, it is fundamental to clarify the difference
between robots and electromechanical devices by defining
what a neurorobot is.
1.2 What is a Neurorobot?
Some cooking machines are commonly called robots by
manufacturers and end-users. However, no one calls a
mixer a robot. This does not depend on machine
complexity: a car is usually more sophisticated than a cooking
machine, but no one considers cars to be robots. In contrast,
clinicians and sometimes neuroscientists often confound
electromechanical devices with robots [
The word ‘‘robot’’ first appeared in 1921 in a science
fiction play titled R.U.R. (Rossum’s Universal Robots)
written by the Czech author Karel Capek. It derives from
the Czech word ‘‘robota’’, meaning hard workers [
The robots invented by Capek were not robots in the
popularly understood sense of mechanical devices; instead,
they were assembled biological organisms. However, the
term has since come to signify primarily electromechanical
devices (often humanoid) endowed with artificial
intelligence and able to perform a variety of functions, partly
through programming and partly through their own ability
to act autonomously . According to that, the Robot
Institute of America defined a robot as ‘‘a programmable,
multi-functional manipulator designed to move material,
parts or specialized devices through variable programmed
motions for the performance of a variety of tasks’’ [
Neurorobotics refers to the branch of science combining
neuroscience, robotics, and artificial intelligence. It hence
refers to all robots developed for interacting with or for
emulating the nervous system of humans or other animals.
A neurorobot can be developed for clinical purposes, for
example neurorehabilitation or neurosurgery, or for
studying the nervous system by emulating its properties, as
it occurs for example in the walking robots based on central
pattern generators [
As mentioned above, a robot should be capable of
performing a variety of tasks. This adaptability is based on its
on-board sensors, the signals of which are processed by
artificial intelligence to change the behaviour of the robot.
Hence, the fundamental point differentiating robots from
electromechanical devices is the adaptability of their
operation. In neurorehabilitation, this differentiation has
often been considered as picky, and robots and
electromechanical devices are often grouped together during
analyses of their efficacy [
]. Treadmills with body
weight support and other devices such as Gait Trainer
(Reha-Stim) should be defined as electromechanical
devices, because, once the physiotherapist has fixed their
parameters, they are not capable of autonomously adapting
them during operation. Conversely, other devices
developed for walking recovery, such as Lokomat (Hocoma,
Volketswil, Switzerland), can be defined as robots since
they use sensors to adapt their functioning to the patient’s
performance (e.g., Lokomat has a position control mode for
applying an assistance-as-needed guidance force to the
1.3 Features of Neurorehabilitation Robots
Many neurorehabilitation approaches and techniques have
been developed to restore neuromotor function, aiming at
the recovery of physiological movement patterns in
patients with neurological pathologies. However, none has
emerged as a gold standard, since it is common opinion
that methods should be specifically tailored for pathologies
and patients [
]. However, a common feature of these
neurorehabilitative approaches is the need for intensive,
repetitive, and task-oriented treatments [
Many authors reported that robots can improve
rehabilitation outcome. In 2008, Wolbrecht et al. [
identified three main desirable features for a controller of
robotaided movement training (see Table 1). One year later,
Morasso et al. [
] re-stated these features, adding the
importance of haptic properties and auto-adaptive
capacities. Then, Belda-Lois et al. [
] suggested four features
for favoring a top-down approach when a robot is used for
post-stroke gait recovery. Finally, Dietz et al. [
reported four main potential advantages of the use of
robots in neurorehabilitation. All these features are listed
in Table 1.
The features indicated by Wolbrecht et al. [
focused on the need of adaptability of neurorobots to
patients’ abilities. Morasso et al. [
] added that a robot
must have also haptic properties and some intelligent
capabilities related to an adaptive assist-as-needed
approach. Both studies highlighted the importance of a
high mechanical compliance, i.e., the need of having a
robot with low-stiffness control. A stiff position controller,
such as that of industrial robots, can move limbs along the
desired trajectories, limiting errors. However, such a
controller impedes error-based learning, which is an essential
component of motor re-learning [
]. Furthermore, a
lowstiffness robot is potentially less dangerous than a
highstiffness robot during interaction with the patient [
other studies [
] focused on the importance of
intensive (for patients, not therapists) and repeatable exercises.
Both pointed out the possibility of exploiting robot sensors
not only to adapt to the patient’s performance, but also to
provide biofeedback to the patient (increasing his/her
motivation and hence participation in rehabilitation), and
feedback to therapists and clinicians on patient progress.
Neurorobots have the potential for accurate assessment
of motor function in order to assess the patient status, to
measure therapy progress, or to give the patient and
therapist real-time feedback on movement performance [
This approach has been proposed in some recent studies.
Kinematic robotic measures, especially those related to
range of motion, have recently been indicated as useful in
the assessment of motor deficits in reaching movements
] and proprioceptive function of hands [
] and upper
] and lower [
] limbs. Furthermore, kinetic robotic
measures have been reported as useful in the assessment of
upper limb strength [
It should be noted as among these features, effectiveness
is not listed, probably because it is taken for granted when
training is performed in a patient-tailored, intensive,
repetitive, and task-oriented manner; however, this issue
deserves further attention.
1.4 Effectiveness Paradox in Neurorobotics
1.5 Other Barriers Limiting Neurorobotics
Morasso et al. noted a paradox in the assessment of
effectiveness of neurorehabilitation robots [
studies have suggested that robotic treatment should be
highly personalized by setting the robot parameters in order
to exploit the residual capabilities of each patient for
recovering a functional status. This implies that in order to
be effective, robotic treatment cannot be standardized, and
therefore controlled clinical trials in the traditional sense
are impossible, unless aimed at very specific and narrow
groups (implying a small sample size, hence poor statistical
evidence). The contrast between a standardized treatment
(with clear guidelines) allowing the design of a randomized
controlled trial (and of clear rehabilitative programmes)
with an adaptable treatment, tailored for patients’
capabilities, is the core of this effectiveness paradox.
Furthermore, the contrast between standardization and adaptability
is not the only problem in designing a methodologically
rigorous study. Intensive training may increase the risk of
inducing or augmenting spasticity. In addition, the
monotony of the same exercise with identical trajectories
clashes with the need for continuous adaptation of robots to
the changing abilities of patients and with the need for
motivating, rather than boring, exercises. Finally, most
robots help patients in reproducing a movement that
replicates the physiological one, despite the fact that most
severely affected patients have a low possibility of a
It should be noted that these inconsistencies are present
also in conventional neurorehabilitation training. The
scientific bases of neuromotor physiology,
neurorehabilitation, and brain plasticity are still not completely clear.
Neurorehabilitation is still mainly ill-defined, with
competing schools of thought about the best treatment.
This generates another scientific roadblock for
neurorobots. In fact, neither the optimal movement tasks nor the
optimal mechanical inputs are well known. Therefore, the
first problem that a robotics engineer encounters when
setting out to build a robotic therapy device is that there is
still substantial uncertainty as to what exactly the device
should do [
], despite the above-cited general features
suggested in the literature.
Interestingly, the scepticism related to neurorobotics due
to the rather inconclusive evaluation of its efficacy and to
the reported inconsistencies is not mitigated by the
consideration that quite similar evaluations could be
formulated for the variety of human-delivered rehabilitation
]. Thus, the doubts about the use of
neurorobots could be not only attributed to the uncertainty
related to efficacy, but also to some other barriers limiting
their wider adoption in rehabilitative settings.
Other aspects limiting neurorobotics are due to
technological, behavioural, and economic barriers [
economic burden is a potential limit for robot adoption in
neurorehabilitation, although it has been reported that the
long-term use of neurorobots can decrease healthcare
system costs [
]. For example, a single physiotherapist could
manage up to four robots (hence four patients) at the same
]. Masiero et al. [
] quantified the cost of using
NeReBot (a robot for the treatment of post-stroke upper
limb impairment) to be 37 % of the hourly physiotherapy
cost, with benefits that include a reduction in
hospitalization time. This suggests that robotic technology can be a
valuable, and an economically sustainable aid, in the
management of patient rehabilitation. Hesse et al. found a
similar percentage (41 %) under the assumption that the
therapist is needed only at the beginning and end of
therapy, and in particular situations where help is needed [
In general, rigorous studies on the economic sustainability
of robots for neurorehabilitation are very sporadic [
These few studies suggest that robotic therapy leads to a
reduction of costs for the healthcare system, in terms of a
reduction in the hospitalization for each patient, higher
autonomy at discharge, or both. However, as highlighted
by Turchetti et al. [
], an individual hospital could be less
interested than the final payer (e.g., the national or local
healthcare system, the private patient, or the insurance
companies) in these aspects. However, this clearly depends
on the reimbursement regimen and on the agreement
between the parties. In general, uncertainty remains about
the cost-effectiveness of robotic neurorehabilitation [
Technological and behavioural aspects could be related
to the possibility that the expectations of patients and
clinicians about outcomes of a neurorobotic treatment are
too high with regards to the current biomedical engineering
level. These reasons seem conceivable, but raise another
question: why have such expectations not limited other
kinds of medical robot, such as surgical robots? In fact,
although surgical robots were introduced at around the
same time as neurorehabilitation robots, their benefit in
assisting surgery (and especially minimally invasive
surgery) is established. Even in fields with no unequivocal
evidence of the superiority of robot-assisted over
traditional surgery, the popularity and diffusion of robotic
surgery has progressively increased [
]. In the last 25 years,
robots have brought a tremendous improvement to the field
of surgery [
]. Thus, other reasons should be investigated
to deeply understand what is still lacking for
neurorehabilitation robots in order to match the expectations of
patients and clinicians. In this scenario, an irrational aspect
seems to play a fundamental role.
1.6 Fear of Robots
In the play of Capek, robots are initially obedient, and,
when commanded, they perform the required task, by
exactly following human instructions. The robots
eventually escape human control and start a rebellion. This theme
is similar to the Jewish myth of the Golem of Prague (an
animated anthropomorphic being entirely created from
inanimate matter) and is used in many science fiction
works. Could fear actually play a role in the scepticism
In general, studies that used questionnaires to collect the
opinions of users (patients and therapists) of
neurorehabilitation robots reported good usability, comfort,
acceptability, and satisfaction. However, most were feasibility
studies that enrolled healthy subjects [
], fewer than 10
], or lacked a control group undergoing
conventional physiotherapy [
]. Even when a control
group was used, only the satisfaction of experimental
physiotherapy was assessed . Hence, these positive
results should be read with caution, since they were
obtained on a small group of users, often not randomly
assigned to robotic therapy. Furthermore, these results can
generate a bias, since the patients, who accepted to undergo
robotic therapy, could be more trustful with regards to the
use of new technological rehabilitation interventions.
In 2000, Burgar et al. reported their experience in
developing robots for neurorehabilitation, concluding their
work with ‘‘we do not view robots as replacements for
]. However, most of the initial studies on
robots claimed that robotic devices can reduce the number
of therapists and the associated costs needed for
25, 56, 57
] (despite the existence of cases in which
two physiotherapists are required for preparing the most
severely affected patients for robotic neurorehabilitation,
which is typically the case when harnessing the patient on
robots for walking recovery based on body weight support
Furthermore, in terms of control, the patient’s feelings
related to robot use in neurorehabilitation should also be
considered. Bragoni et al. [
] identified the level of
anxiety of patients as a negative prognostic factor for robotic
therapy but not for conventional therapy. In contrast,
patients who saw themselves as the chief causal factor in
managing their recovery showed higher probability of a
better outcome with robotic rehabilitation [
]. This kind
of fear could be due to the sensation that robots are not
considered trustworthy because they lack human feelings,
expertise, and common sense [
]. This is one of the
hardest problems in artificial intelligence and robotics
faced by bioengineers.
2 Three Laws of Neurorobotics
2.1 Three Laws of Robotics
After the play of Capek, robots became iconic, especially
thanks to Isaac Asimov’s stories, and to his compilation ‘‘I,
Robot’’ in 1950 [
]. In a story included in that
compilation and first published in 1942 titled ‘‘Runaround’’,
Asimov invented the three laws of robotics, quoted as being
from the ‘‘Handbook of Robotics, 56th Edition, 2058’’.
These rules are a set of fundamental requirements for the
design and manufacture of intelligent robots. They are
intended to ensure that robots will operate for the benefit of
humanity, rather than becoming a threat to humans. These
laws had a very influential role in subsequent science
fiction works, and became also important with the emergence
of robotics as a scientific discipline [
]. The three laws of
A robot may not injure a human being or, through
inaction, allow a human being to come to harm.
A robot must obey the orders given it by human
beings, except where such orders would conflict with
the First Law.
A robot must protect its own existence, as long as
such protection does not conflict with the First or
These laws define a kind of set of ethic rules for robots
(or for the human programmers of their artificial
intelligence). The hierarchical structure of these laws places at
the first level human health, followed by human will, and
finally robot self-preservation. These laws should not be
considered only as part of science fiction imagery. Their
potential role is so important that they have been
re-analyzed in the current context, in the Editorial of a Special
Issue of Science, entitled ‘‘Robot Ethics’’ [
]. In this
editorial, Sawyer stated that, since the U.S. military is a
major source of funding for robotic research, it is unlikely
that such laws will be integrated in their design. This
argument can be generalized to cover other robotic
industries: the development of artificial intelligence is a
business, and businesses are usually uninterested in ethical
issues. The risk, in the neurorehabilitation field, is that
companies may produce attractive robots without proving
their effectiveness. The potential risks related to the use of
medical robotics deserve attention: harm may occur from
anomalous functioning, or even from normal robot
]. If many of the problems related to neurorobots
are related to fear, risks, and ethical issues, it is probably
time to define a set of rules for neurorobot ethics before
defining their desirable features.
2.2 Three laws of neurorobotics
According to the aforementioned desirable features of a
neurorobot, we have re-formulated the three laws of
robotics into three laws for robotics in neurorehabilitation:
A robot for neurorehabilitation may not injure a
patient or allow a patient to come to harm.
A robot must obey the orders given it by therapists,
except where such orders would conflict with the
A robot must adapt its behavior to patients’ abilities
in a transparent manner as long as this does not
conflict with the First or Second Law.
These laws and their implications are discussed below.
3.1 First Law of Neurorobotics: Need for High
Personal care robots (e.g., mobile servant robots, physical
assistant robots, and person carrier robots) should be
designed in accordance with the international standards
defined by ISO 13482:2014 [
]. In 2014, the International
Organization for Standardization published these criteria
for designing personal care robots, providing the needed
requirements to eliminate or reduce the risks associated
with the use of medial robots to an acceptable level. ISO
13482:2014 is more specific for personal care robots,
including neurorobots, than the previous ISO14971:2000
]. ISO 13482:2014 can be considered to be in line with
the first law of Asimov, with ‘‘harm’’ referring to that to the
patient. Datteri [
], in a review about responsibility in
using medical robots (including surgery and diagnostic
robots, neurorehabilitation robots, robotic prostheses, and
even next-generation personal assistance robots), stated
that these devices operate in close proximity or direct
physical contact with humans, manipulate instruments
inside the patient’s body or directly move user’s impaired
limbs, and have invasive or non-invasive connections with
the human nervous system. They can hence contribute to
improving the precision of medical treatments, relieving
therapists of tasks that require considerable accuracy and
physical effort, and improving the quality of life of patients
]. Nevertheless, they also may threaten the physical
integrity of patients, not only through harmful events
caused by anomalous behaviours (e.g., in surgery), but
even through normal operation [
]. This can typically
occur for neurorehabilitation robots whose efficacy has not
been proven [
]. Datteri’s review gives the example of
Lokomat, showing that, despite its diffusion in many
rehabilitation centers, there is neither well-supported
experimental nor theoretical evidence that Lokomat-based
therapies are at least as beneficial as conventional
therapies. Instead, the review gives examples of studies that
showed that Lokomat reproduces abnormal and
nonphysiological gait patterns due to the restriction of pelvis
movement, altering lower limb joint kinematics [
muscle activations [
]. This limitation has recently been
overcome in Lokomat Pro (Hocoma) by the addition of an
optional module that allows lateral translation and
transverse rotation of the pelvis, aiming at a more physiological
movement. However, it is still unclear if training based on
physiological movement is the optimal solution for patients
severely affected and probably unable to completely
recover physiological patterns. In fact, recovery of
autonomy in walking should be the objective of robotic gait
rehabilitation, where recovery of physiological gait
patterns is not mandatory.
Neurorobots should be safe not only in terms of
movement, but also from other medical points of view. For
example, despite the variety of gait patterns, robotic gait
training performed with body weight support has only
recently been proven safe for training intensive walking in
non-autonomous ambulatory patients with subacute stroke.
The reason is that the cardio-respiratory demand is lower
than that in conventional walk training performed
]. Interestingly, the authors found the opposite
result for healthy subjects: overground walking was less
demanding than robotic walking. They suggested that this
could have been because the robot imposes non-natural
trajectories, which force subjects to activate non-natural
sensorimotor walking patterns.
We would like to enlarge the meaning of ‘‘harm’’ to all
possible damage to patients. Time spent on an ineffective,
slightly effective, or even detrimental robot should be
considered as damage, because the patient could spend the
same time in a more effective treatment. Hence, the first
law implies that robot usage should be at least as safe and
effective as other treatments, meaning that it should have a
higher benefit-risk ratio than that of human-administered
treatments. This ratio should be evaluated before
commercialization of the device, and not afterwards, as is often
But how can a robot be effective in the light of the cited
effectiveness paradox and in the absence of a clear scientific
background? Firstly, it is probably time to delay the
commercial launch of neurorobots until a deep examination of
their potential effectiveness is conducted, adopting an
approach more similar to that used in other medical or
engineering disciplines. For example, specific rules are
defined for clinical trials prior to drug commercialization
(Table 2). These trials require Phase I, (commonly
performed in the producer laboratories), followed by Phases II
and III (performed in independent hospitals), before
commercialization can occur. Further, Phase IV follows in
clinical or daily living settings. Dobkin redefined these phases
for motor rehabilitation treatments [
] (refer to Table 2),
and we suggest that a similar roadmap should be followed by
companies before commercialization of neurorobots (that
should occur only after an equivalent Phase III).
Furthermore, for neurorehabilitation robots, there is still
a lack of clear information about how to administer robotic
therapy, proper use, treatment duration and frequency,
precautions, possible side effects, etc. However, the
effectiveness of a treatment (including that with a
neurorehabilitation robot) depends on the patient
characteristics (e.g., type and severity of disease, presence of specific
], on the duration and frequency of sessions to
administer, and on the correct phase of rehabilitation at
which the therapy should be administered [
example, Morone et al. reported that patients with more
severe impairments in the motor leg benefited more from
robotic-assisted therapy than did patients with greater
voluntary motor function in the affected limb, who can
perform intensive and less constrained training in
conventional therapy [
]. Unfortunately, neurorobot
handbooks are at the moment still similar to generic
commercial pamphlets, far from drug information sheets.
3.2 Second Law of Neurorobotics: Tool for Therapists
Some therapists see a robot as a possible substitute for their
work. Morasso et al. thus titled their review on robots for
rehabilitation ‘‘Desirable features of a ‘humanoid’
]. Hidler et al. emphasized that the goal of
introducing robots into rehabilitation hospitals is not to
replace therapists, but rather to complement existing
treatment options [
]. Nevertheless, it is reasonable to
believe that the reduction of healthcare costs is at least one
of the main motives driving research in neurorobotics [
given that many studies have reported that robots may
reduce the cost of rehabilitation by reducing the number of
required therapists [
25, 56, 57
The higher popularity of neurosurgery robots compared
to neurorehabilitation robots is thus likely due to the fact
that the former do not replace the surgeon, but aid him.
Similarly, a robot for rehabilitation should not be
considered as a standing-alone rehabilitation device [
], but a
tool in the hands of therapists, giving them more precise
movements, more intensive, repeatable, or adaptable
patterns, according to the therapists’ expertise, and relieving
them from fatigue. The therapist should therefore be
included in the loop, in order to drive the symbiotic
equilibrium between robot and patient towards an optimum, by
dialoguing with the patient, motivating them, and getting
verbal feedback on fatigue, pain, and emotional stress
(parameters difficult to monitor with sensors) [
Recently, the need for a therapist as motivator to avoid the
patient having a passive role during robotic therapy has
been overcome by a top-down approach of robots
combined with stimulating biofeedback, video-game-based
therapy, and even brain-computer interfaces [
However, a therapist should play a key role in terms of
robotic therapy administration, such as robot parameter
adjustments, avoiding harmful patient compensation
strategies, identification of the trade-off between
challenging tasks that help rehabilitation and those that
To this end, we propose to extend the loop proposed by
Morasso et al. [
] to include the therapist (see Fig. 1). In
our opinion, the desired reduction of costs for the
healthcare system can be obtained not by reducing the number of
therapists, but increasing the efficacy of rehabilitation,
reducing the length of stay in rehabilitative hospitals, and
releasing more autonomous patients with a consequent
reduction of home care costs.
The proposed second law of neurorobotics, making the
robot perfectly obedient to the therapists’ requests, may
seem obvious, but it is not. Besides the above-mentioned
problems related to non-physiological gait patterns in
Lokomat-based therapy [
], another example of robot
‘‘disobedience’’ is the discrepancy between the desired and
actual values of some parameters of the electromechanical
Gait Trainer (as highlighted in [
]). The effective
percentage of body weight supported by the machine is
different from that selected in the initial static condition, since
the machine does not take into account the changes that
occur in the patient capacity to support their own weight
during training. Furthermore, the authors highlighted that
for Gait Trainer, the defined selector of walking speed is
actually a selector of step duration, and that the reported
speed coincides with the real one only if the maximum step
length has been also selected.
Robots should ‘‘disobey’’ clinicians’ orders only if their
sensors indicate that such orders lead to a potential risk for
the patient. This highlights the importance of sensors,
which is at the base of the adaptability and autonomy of
any robotic system [
]. In contrast, an electromechanical
device is not required to detect a potentially dangerous
choice by therapists due to wrong parameter tuning.
The presence of a therapist in the loop (Fig. 1) allows
human control of the device, but the robot’s artificial
intelligence should not be limited to the safety control of
human decisions. During rehabilitation, there are many
parameters to calibrate, tune, and adapt. Firstly, the
clinician should always consider the effects of a parameter
change on other parameters. For example, to increase speed
during overground walking, a subject can reduce step
duration, increase step length, or both (usually at the same
time). In Lokomat-based training, when a therapist
increases the patient’s walking speed, they are actually
reducing the step duration without altering the step length,
since this parameter depends on the sagittal range of hip
motion; such changes in that hip range of motion need a
manual adjustment by the therapist. The handbook of
] suggests that therapists should consider the
following points when increasing speed: (1) manually
adapt step length acting on hip range of motion controller
(the wider is the hip movement, the longer is the step); (2)
adjust the synchronization between treadmill and
exoskeleton speed (automatic setting is also possible); (3)
adjust the hip offset (not only range); (4) take into account
that foot impact could increase, and hence increase the load
on the joints; (5) check the quality of the movement that
may be affected by the change. This highlights how many
parameters are related to a simple change of speed in a
robot for gait training. Furthermore, speed is a parameter
with a very clear physiological meaning. More problems
could occur for a parameter for which it is not so easy to
understand its role, such as guidance force.
Robot artificial intelligence should be capable of
automatically performing all the control changes required by
the therapist, while providing them with a clear
quantitative overview of all these changes. The adoption of robotic
technologies for helping patients and therapists and
quantitatively evaluating patient recovery is the main issue of
European projects such as MAAT (‘‘Multimodal interfaces
to improve therapeutic outcomes in robot-assisted
rehabilitation’’, www.echord.info/wikis/website/maat) and
SYMBITRON (‘‘Symbiotic man–machine interactions in
wearable exoskeletons to enhance mobility for
paraplegics’’, www.symbitron.eu). These projects include the
patient in a symbiotic loop with the robot, similarly to what
we suggest in Fig. 1. Then, the therapist should simply be
required to qualitatively control patient performance under
the new conditions.
Summarising these concepts: a new generation of
human–machine interfaces integrated in neurorobots
should be developed, in which the therapist’s commands at
the macro level can be translated in micro changes
autonomously by the robot, which should inform the
therapist of these changes. However, there are no easy
ways to assess algorithmically whether the mutual
patientrobot adaptation is the optimal one for favouring the
neuromotor recovery [
]. For this reason, the therapist should
be kept in the loop. In contrast with the robot, the therapist
has a qualitative but natural access to the health status of
the patient. For instance, they have detailed feedback of
feelings and sensations by dialoguing with the patient.
Most studies and reviews about robots for
neurorehabilitation have focused on their effectiveness, but have found
inconsistent results. Little attention has been given to robot
ethics, probably because artificial intelligence is still
primitive. However, data shows that patients and therapists
are somewhat afraid of robots. Although we did not suggest
new technical solutions, in this review, we described the
state of the art of robots for neurorehabilitation, and
suggested a set of rules, which are a re-formulation of
Asimov’s three laws of robotics. We indicated the need for
these laws with many examples. The proposed three laws
of neurorobotics highlight the ethical need to prove a
robot’s effectiveness before commercialization, as well as
the desirable features that neurorobots should have.
Furthermore, we highlighted the need for including the
therapist in the loop between patient and robot. Finally, we
suggested that neurorobots can be a valuable tool in
therapists’ hands, helping them not only in repetitive and
intensive patient mobilization, but also providing
quantitative information about a patient’s deficits, residual
abilities, and functional recovery. We think that these three
laws should be considered from the first stages of
neurorobot design. They may bring together engineers and
clinicians for the development of a new, effective
generation of robots for neurorehabilitation.
Compliance with Ethical Standards
Conflict of interest The authors declare that they have no conflict
Open Access This article is distributed under the terms of the
Creative Commons Attribution 4.0 International License (http://crea
tivecommons.org/licenses/by/4.0/), which permits unrestricted use,
distribution, and reproduction in any medium, provided you give
appropriate credit to the original author(s) and the source, provide a
link to the Creative Commons license, and indicate if changes were
1. Krebs , H. I. , Hogan , N. , Aisen , M. L. , & Volpe , B. T. ( 1998 ). Robot-aided neurorehabilitation . IEEE Transactions on Rehabilitation Engineering , 6 ( 1 ), 75 - 87 .
2. Gosine , R. G. , Harwin , W. S. , Furby , L. J., & Jackson , R. D. ( 1989 ). An intelligent end-effector for a rehabilitation robot . Journal of Medical Engineering & Technology , 13 ( 1-2 ), 37 - 43 .
3. Preising , B. , Hsia , T. C. , & Mittelstadt , B. ( 1991 ). A literature review: Robots in medicine . IEEE Engineering in Medicine and Biology Magazine , 10 ( 2 ), 13 - 22 .
4. Van Vliet , P. , & Wing , A. M. ( 1991 ). A new challenge-robotics in the rehabilitation of the neurologically motor impaired . Physical Therapy , 71 ( 1 ), 39 - 47 .
5. Aisen , M. L. , Krebs , H. I. , Hogan , N. , McDowell , F. , & Volpe , B. T. ( 1997 ). The effect of robot-assisted therapy and rehabilitative training on motor recovery following stroke . Archives of Neurology , 54 ( 4 ), 443 - 446 .
6. Hesse , S. , Schmidt , H. , Werner , C. , & Bardeleben , A. ( 2003 ). Upper and lower extremity robotic devices for rehabilitation and for studying motor control . Current Opinion in Neurology , 16 ( 6 ), 705 - 710 .
7. Veneman , J. F. , Kruidhof , R. , Hekman , E. E. , Ekkelenkamp , R., Van Asseldonk , E. H. , & Van der Kooij , H. ( 2007 ). Design and evaluation of the LOPES exoskeleton robot for interactive gait rehabilitation . IEEE Transactions on Neural Systems and Rehabilitation Engineering , 15 ( 3 ), 379 - 386 .
8. Hidler , J. , Nichols , D. , Pelliccio , M. , Brady , K. , Campbell , D. D. , Kahn , J. H. , et al. ( 2009 ). Multicenter randomized clinical trial evaluating the effectiveness of the Lokomat in subacute stroke . Neurorehabilitation and Neural Repair , 23 , 5 - 13 .
9. Husemann , B. , Mu¨ller, F. , Krewer , C. , Heller , S. , & Koenig , E. ( 2007 ). Effects of locomotion training with assistance of a robotdriven gait orthosis in hemiparetic patients after stroke: A randomized controlled pilot study . Stroke , 38 , 349 - 354 .
10. Pohl , M. , Werner , C. , Holzgraefe , M. , Kroczek , G. , Mehrholz , J. , Wingendorf , I. , et al. ( 2007 ). Repetitive locomotor training and physiotherapy improve walking and basic activities of daily living after stroke: A single-blind, randomized multicentre trial (DEutsche GAngtrainerStudie , DEGAS). Clinical Rehabilitation , 21 , 17 - 27 .
11. Tong , R. K. , Ng , M. F. , & Li , L. S. ( 2006 ). Effectiveness of gait training using an electromechanical gait trainer, with and without functional electric stimulation, in subacute stroke: A randomized controlled trial . Archives of Physical & Medicine Rehabilitation , 87 , 1298 - 1304 .
12. Lo , A. C. , Guarino , P. D. , Richards , L. G. , Haselkorn , J. K. , Wittenberg , G. F. , & Federman , D. G. ( 2010 ). Robot-assisted therapy for long-term upper-limb impairment after stroke . The New England Journal of Medicine , 362 ( 19 ), 1772 - 1783 .
13. Klamroth-Marganska , V. , Blanco , J. , Campen , K. , Curt , A. , Dietz , V. , Ettlin , T. , et al. ( 2014 ). Three-dimensional, taskspecific robot therapy of the arm after stroke: A multicentre, parallel-group randomised trial . Lancet Neurology , 13 ( 2 ), 159 - 166 .
14. Mehrholz , J. , Platz , T. , Kugler , J. , Pohl , M. ( 2008 ). Electromechanical and robot-assisted arm training for improving arm function and activities of daily living after stroke . The Cochrane Database of Systematic Reviews , 8 ( 4 ): CD006876 .
15. Mehrholz , J. , Ha¨drich, A. , Platz , T. , Kugler , J. , Pohl , M. ( 2012 ). Electromechanical and robot-assisted arm training for improving generic activities of daily living, arm function, and arm muscle strength after stroke . The Cochrane Database of Systematic Reviews , 6 : CD006876 .
16. Mehrholz , J. , Werner , C. , Kugler , J. , Pohl , M. ( 2007 ). Electromechanical-assisted training for walking after stroke . The Cochrane Database of Systematic Reviews . 4 : CD006185 .
17. Mehrholz , J. , Elsner , B. , Werner , C. , Kugler , J. , Pohl , M. ( 2013 ). Electromechanical-assisted training for walking after stroke . The Cochrane Database of Systematic Reviews . 7 : CD006185 .
18. Mehrholz , J. , Pohl , M. , Elsner , B. ( 2014 ). Treadmill training and body weight support for walking after stroke . The Cochrane Database of Systematic Reviews . 1 : CD002840 .
19. Iosa , M. , Morone , G. , Fusco , A. , Bragoni , M. , Coiro , P. , Multari , M. , et al. ( 2012 ). Seven capital devices for the future of stroke rehabilitation . Stroke Research & Treatment , 2012 , 187965 .
20. Morasso , P. , Casadio , M. , Giannoni , P. , Masia , L. , Sanguineti , V. , Squeri , et al. ( 2009 ). Desirable features of a ''humanoid'' robottherapist . Conference proceedings: Annual International Conference of the IEEE Engineering in Medicine and Biology Society , 2009 : 2418 - 2421 .
21. Morone , G. , Bragoni , M. , Iosa , M. , De Angelis , D. , Venturiero , V. , Coiro , P. , et al. ( 2011 ). Who may benefit from robotic-assisted gait training? A randomized clinical trial in patients with subacute stroke . Neurorehabilitation & Neural Repair , 25 ( 7 ), 636 - 644 .
22. Morone , G. , Iosa , M. , Bragoni , M. , De Angelis , D. , Venturiero , V. , Coiro , P. , et al. ( 2012 ). Who may have durable benefit from robotic gait training?: A 2-year follow-up randomized controlled trial in patients with subacute stroke . Stroke , 43 ( 4 ), 1140 - 1142 .
23. Bragoni , M. , Broccoli , M. , Iosa , M. , Morone , G. , De Angelis , D. , Venturiero , V. , et al. ( 2013 ). Influence of psychologic features on rehabilitation outcomes in patients with subacute stroke trained with robotic-aided walking therapy . American Journal of Physical & Medicine Rehabilitation , 92 ( 10 Suppl 2 ), e16 - e25 .
24. Iosa , M. , Morone , G. , Bragoni , M. , De Angelis , D. , Venturiero , V. , Coiro , P. , et al. ( 2011 ). Driving electromechanically assisted Gait Trainer for people with stroke . Journal of Rehabilitation Research and Development , 48 ( 2 ), 135 - 146 .
25. Masiero , S. , Poli , P. , Rosati , G. , Zanotto , D. , Iosa , M. , Paolucci , S. , et al. ( 2014 ). The value of robotic systems in stroke rehabilitation . Expert Review of Medical Devices , 11 ( 2 ), 187 - 198 .
26. Roberts , A. ( 2006 ). The history of science fiction . New York, NY: Palgrave MacMillan.
27. Booker , K. M. ( 2015 ). Historical dictionary of science fiction in literature . Lanham, Maryland: Rowman & Littlefield.
28. Xie , Ming. ( 2003 ). Fundamental of robotics: Linking perception to action . Singapore: World Scientific.
29. Ijspeert , A. J. , Crespi , A. , Ryczko , D. , & Cabelguen , J. M. ( 2007 ). From swimming to walking with a salamander robot driven by a spinal cord model . Science , 315 , 1416 - 1420 .
30. Belda-Lois , J. M. , Mena-del Horno , S. , Bermejo-Bosch , I. , Moreno , J. C. , Pons , J. L. , Farina , D. , et al. ( 2011 ). Rehabilitation of gait after stroke: A review towards a top-down approach . Journal of Neuroengineering & Rehabilitation , 8 , 66 .
31. Wolbrecht , E. T. , Chan , V. , Reinkensmeyer , D. J. , & Bobrow , J. E. ( 2008 ). Optimizing compliant, model-based robotic assistance to promote neurorehabilitation . IEEE Transactions on Neural Systems and Rehabilitation Engineering , 16 ( 3 ), 286 - 297 .
32. Dietz , V. , Nef , T. , & Rymer , W. Z. ( 2012 ). Neurorehabilitation technology . London, UK: Springer.
33. Keller, U., Scho¨lch, S. , Albisser , U. , Rudhe , C. , Curt , A. , Riener , R. , & Klamroth-Marganska , V. ( 2015 ). Robot-assisted arm assessments in spinal cord injured patients: A consideration of concept study . PLoS ONE , 10 ( 5 ), e0126948 .
34. Otaka , E. , Otaka , Y. , Kasuga , S. , Nishimoto , A. , Yamazaki , K. , Kawakami , M. , et al. ( 2015 ). Clinical usefulness and validity of robotic measures of reaching movement in hemiparetic stroke patients . Journal of Neuroengineering & Rehabilitation , 12 , 66 .
35. Ingemanson , M. L. , Rowe , J. B. , Chan , V. , Wolbrecht , E. T. , Cramer , S. C. , & Reinkensmeyer , D. J. ( 2016 ). Use of a robotic device to measure age-related decline in finger proprioception . Experimental Brain Research , 234 ( 1 ), 83 - 93 .
36. Cappello , L. , Elangovan , N. , Contu , S. , Khosravani , S. , Konczak , J. , & Masia , L. ( 2015 ). Robot-aided assessment of wrist proprioception . Frontiers in Human Neuroscience , 9 , 198 .
37. Domingo , A. , & Lam , T. ( 2014 ). Reliability and validity of using the Lokomat to assess lower limb joint position sense in people with incomplete spinal cord injury . Journal of Neuroengineering & Rehabilitation , 11 , 167 .
38. Van Der Loos , H. F. M. , & Reinkensmeyer , D. J. ( 2008 ). Rehabilitation and health care robotics . In B. Siciliano & O. Khatib (Eds.) , Springer Handbook of Robotics (pp. 1223 - 1251 ). Berlin: Springer.
39. Turchetti , G. , Vitiello , N. , Trieste , L. , Romiti , S. , Geisler , E. , & Micera , S. ( 2014 ). Why effectiveness of robot-mediated neurorehabilitation does not necessarily influence its adoption . IEEE Reviews in Biomedical Engineering , 7 , 143 - 153 .
40. Masiero , S. , Poli , P. , Armani , M. , Ferlini , G. , Rizzello , R. , & Rosati , G. ( 2014 ). Robotic upper limb rehabilitation after acute stroke by NeReBot: Evaluation of treatment costs . Biomed Research International, 2014 , 265634 .
41. Hesse , S. , Heß , A. , Werner , C. , Kabbert , N. , & Buschfort , R. ( 2014 ). Effect on arm function and cost of robot-assisted group therapy in subacute patients with stroke and a moderately to severely affected arm: A randomized controlled trial . Clinical Rehabilitation , 28 ( 7 ), 637 - 647 .
42. Turchetti , G. , Vitiello , N. , Trieste , L. , Romiti , S. , Geisler , E. , & Micera , S. ( 2014 ). Why effectiveness of robot-mediated neurorehabilitation does not necessarily influence its adoption . IEEE Reviews in Biomedical Engineering , 7 , 143 - 153 .
43. Wagner , T. H. , Lo , A. C. , Peduzzi , P. , Bravata , D. M. , Huang , G. D. , Krebs , H. I. , et al. ( 2011 ). An economic analysis of robotassisted therapy for long-term upper-limb impairment after stroke . Stroke , 42 ( 9 ), 2630 - 2632 .
44. Ng , A. T. , & Tam , P. C. ( 2014 ). Current status of robot-assisted surgery . Hong Kong Medical Journal , 20 ( 3 ), 241 - 250 .
45. Spetzger , U. , Von Schilling , A. , Winkler , G. , Wahrburg , J. , & Ko¨nig, A. ( 2013 ). The past, present and future of minimally invasive spine surgery: A review and speculative outlook . Minim Invasive Therapy & Allied Technologies , 22 ( 4 ), 227 - 241 .
46. Ama , A. J. , Gil-Agudo , A. , Pons , J. L. , & Moreno , J. C. ( 2014 ). Hybrid FES-robot cooperative control of ambulatory gait rehabilitation exoskeleton . Journal of Neuroengineering & Rehabilitation , 11 , 27 .
47. Sale , P. , Lombardi , V. , & Franceschini , M. ( 2012 ). Hand robotics rehabilitation: Feasibility and preliminary results of a robotic treatment in patients with hemiparesis . Stroke Research & Treatment , 2012 , 820931 .
48. Vanmulken , D. A. , Spooren , A. I. , Bongers , H. M. , & Seelen , H. A. ( 2015 ). Robot-assisted task-oriented upper extremity skill training in cervical spinal cord injury: A feasibility study . Spinal Cord , 53 ( 7 ), 547 - 551 .
49. Park , W. , Jeong , W. , Kwon , G. H. , Kim , Y. H. , & Kim , L. ( 2013 ). A rehabilitation device to improve the hand grasp function of stroke patients using a patient-driven approach . IEEE International Conference on Rehabilitation Robotics , 2013 , 6650482 .
50. Jardo´n, A. , Gil , A ´ . M., de la Pen˜a, A. I. , Monje , C. A. , & Balaguer , C. ( 2011 ). Usability assessment of ASIBOT: A portable robot to aid patients with spinal cord injury . Disability & Rehabilitation. Assist Technology , 6 ( 4 ), 320 - 330 .
51. McCabe , J. P. , Dohring , M. E. , Marsolais , E. B. , Rogers , J. , Burdsall , R. , Roenigk , K. , et al. ( 2008 ). Feasibility of combining gait robot and multichannel functional electrical stimulation with intramuscular electrodes . Journal of Rehabilitation Research and Development , 45 ( 7 ), 997 - 1006 .
52. Bovolenta , F. , Sale , P. , Dall'Armi , V. , Clerici , P. , & Franceschini , M. ( 2011 ). Robot-aided therapy for upper limbs in patients with stroke-related lesions. Brief report of a clinical experience . Journal of Neuroengineering & Rehabilitation , 8 , 18 .
53. Treger , I. , Faran , S. , & Ring , H. ( 2008 ). Robot-assisted therapy for neuromuscular training of sub-acute stroke patients. A feasibility study . European Journal of Physical & Rehabilitation Medicine , 44 ( 4 ), 431 - 435 .
54. Masiero , S. , Celia , A. , Rosati , G. , & Armani , M. ( 2007 ). Roboticassisted rehabilitation of the upper limb after acute stroke . Archives of Physical Medicine and Rehabilitation , 88 ( 2 ), 142 - 149 .
55. Burgar , C. G. , Lum , P. S. , Shor , P. C. , & Machiel Van der Loos , H. F. ( 2000 ). Development of robots for rehabilitation therapy: The Palo Alto VA/Stanford experience . Journal of Rehabilitation Research and Development , 37 ( 6 ), 663 - 673 .
56. Hidler , J. , Nichols , D. , Pelliccio , M. , & Brady , K. ( 2005 ). Advances in the understanding and treatment of stroke impairment using robotic devices . Topics in Stroke Rehabilitation , 12 ( 2 ), 22 - 35 .
57. Datteri , E. ( 2013 ). Predicting the long-term effects of humanrobot interaction: A reflection on responsibility in medical robotics . Science and Engineering Ethics , 19 ( 1 ), 139 - 160 .
58. Asimov , I. ( 1951 ). I, Robot . New York, NY: Gnome Press.
59. Siciliano , B. ( 2008 ). Handbook of robotics . Oussama Khatib: Springer.
60. Sawyer , R. J. ( 2007 ). Robot ethics . Science , 318 ( 5853 ), 1037 .
61. International Organization for Standardization. ( 2014 ). Robots and robotic devices: Safety requirements for personal care robots . ISO , 13482 , 2014 .
62. International Organization for Standardization. ( 2000 ). Medical devices: Application of risk management to medical devices . ISO , 14971 , 2000 .
63. Datteri , E. , & Tamburrini , G. ( 2009 ). Ethical reflections on health care robotics . In R. Capurro & M. Nagenborg (Eds.), Ethics and robotics (pp. 35 - 48 ). Amsterdam: IOS Press/AKA.
64. Regnaux , J. P. , Saremi , K. , Marehbian , J. , Bussel , B. , & Dobkin , B. H. ( 2008 ). An accelerometry-based comparison of 2 robotic assistive devices for treadmill training of gait . Neurorehabilitation & Neural Repair , 22 ( 4 ), 348 - 354 .
65. Hidler , J.M. , & Wall , A.E. ( 2005 ). Alterations in muscle activation patterns during robotic-assisted walking . Clinical Biomechanics (Bristol, Avon) , 20 ( 2 ), 184 - 193 .
66. Delussu , A. S. , Morone , G. , Iosa , M. , Bragoni , M. , Traballesi , M. , & Paolucci , S. ( 2014 ). Physiological responses and energy cost of walking on the Gait Trainer with and without body weight support in subacute stroke patients . Journal of Neuroengineering & Rehabilitation , 11 , 54 .
67. Dobkin , B. H. ( 2009 ). Progressive Staging of Pilot Studies to Improve Phase III Trials for Motor Interventions . Neurorehabilitation & Neural Repair , 23 ( 3 ), 197 - 206 .
68. Morone , G. , Masiero , S. , Werner , C. , & Paolucci , S. ( 2014 ). Advances in neuromotor stroke rehabilitation . Biomed Research International, 2014 , 236043 .
69. Hocoma . Lokomat User Script . http://knowledge.hocoma.com/ fileadmin/user_upload/training_material/lokomat/Lokomat_User_ Script_EN_150511.pdf.