Ahead of his keynote at the IFESS annual congress in Lecco, Milos R. Popovic argues that locomotion is fundamentally a matter of balance, not stepping, and that current robotic rehabilitation systems can help patients stand but cannot yet teach the nervous system to truly regain postural control. Functional electrical stimulation, he explains, has the potential to close this gap, though the technology has not yet reached the maturity needed for widespread clinical use.
After more than two decades of research, Milos R. Popovic argues that functional electrical stimulation has already demonstrated what it can achieve – particularly in upper-limb rehabilitation after stroke. The challenge now is not only technological, but also clinical, organizational and cultural: ensuring that effective solutions are integrated into rehabilitation pathways.
Popovic is Director of the KITE Research Institute at the Toronto Rehabilitation Institute–University Health Network and of the Institute of Biomedical Engineering at the University of Toronto. He will be one of the keynote speakers at the IFESS Congress, taking place in Lecco from 2 to 4 September.
In this interview, he discusses the potential of FES, the limits of current robotic rehabilitation, the role of error-driven motor learning and the responsibility of research centers to translate scientific knowledge into solutions with real clinical and societal impact.
You have spent more than twenty years developing functional electrical stimulation technologies. Where are we today?
I would say that, for upper-limb rehabilitation after stroke, we know what to do. Over the past twenty years we have developed electrical stimulation systems that are comfortable enough for everyday clinical use. We designed dry electrodes that do not require conductive gel, integrated them into wearable garments, and developed stimulation protocols capable of restoring hand function through neuromodulation principles. The technology has been used in more than one thousand patients. According to our clinical experience, every patient showed improvement, and I am not talking about marginal improvements. These were substantial functional gains.
This technology became MyndMove. If it works, what is preventing its wider adoption in clinical practice?
The challenge has been convincing healthcare systems to use it. Several years ago, influential academic publications suggested that patients whose hand function falls below a certain threshold after stroke have very limited recovery potential. Those conclusions influenced healthcare policies. Governments reduced funding for long-term upper limb rehabilitation and shortened hospital stays. In North America many stroke patients now remain in rehabilitation for only two or three weeks. Europe is gradually following the same direction.
Why do you think this is a problem?
Because it is simply inconsistent with what we observe clinically. If patients receive one hour of FES-based therapy every day, they continue to improve. Naturally, recovery is faster when therapy starts immediately after stroke, but improvement is still possible many years later. We have seen meaningful recovery even decades after the original injury.
Can you quantify these improvements?
We usually use the Fugl-Meyer Assessment, which is probably the most widely accepted clinical scale for evaluating motor recovery after stroke. Most robotic systems and conventional interventions report improvements of approximately five up to six points on the upper-limb section of the Fugl-Meyer scale. A change of about six points is generally considered the minimum clinically meaningful difference. With MyndMove, we have observed improvements of approximately twenty-four points. Patients do not need to retain any voluntary hand movement before treatment. They may start with virtually no functional hand activity and still achieve gains approaching half of the entire scale. That changes people’s lives.
So far, we have focused on upper-limb rehabilitation. When we turn to locomotion, what are the main challenges?
Most robotic systems are designed around locomotion itself, while the real problem is balance. Everyone wants to teach patients how to walk, but walking is not the first motor skill humans learn. Look at how children develop. They first crawl, strengthening their trunk muscles. Then they pull themselves up, learning to balance while holding onto furniture. Only after mastering postural control does walking emerge naturally. Locomotion is essentially dynamic balance.
In rehabilitation, however, we often reverse this developmental sequence. We train reciprocal stepping patterns without first restoring balance control. From my perspective, this is the fundamental mistake.
Why is restoring balance so important from a functional point of view?
If a person with spinal cord injury can stand up from a wheelchair, maintain balance while washing their hands, or transfer independently from the wheelchair to the toilet, that has a much greater impact on daily life than simply producing a walking pattern.
Standing and balancing require simultaneous control of trunk muscles, lower-limb muscles and postural adjustments. Robotics can help patients stand, but it cannot yet teach the nervous system how to regain true balance because a vast number of muscles, muscle fibers and postural adjustments must be coordinated simultaneously. Functional electrical stimulation has the potential to address this problem, although the technology has not yet reached the level of maturity required for widespread clinical implementation. Until we solve balance, we will not truly solve locomotion.
Why do you believe current robotic rehabilitation systems have intrinsic limitations?
Because every human being moves differently. Your movement kinematics are different from mine. If I program a robot using my own movement patterns, those patterns will not necessarily match how your brain naturally organizes movement. There is a mismatch- a dissonance – between the movement generated by the robot and the internal motor representation in your brain.
This is why patients often disengage during robotic therapy. After a few minutes they stop paying attention. They look around the room, watch the physiotherapist, or stare out of the window. They are no longer actively involved in the movement because the movement itself is not being generated by their own nervous system. We invested millions of dollars trying to make robotic systems more adaptive and engaging, but we eventually realized that the limitation is much deeper.
You argue that movement learning is fundamentally based on prediction and error. Could you explain this concept?
Every movement we perform is slightly different from the previous one, and this is not because the nervous system is imperfect. Many neuroscientists have traditionally argued that movement variability simply reflects biological noise. I completely disagree. Our body is constantly changing. I may put on a backpack, wear heavier clothes, become fatigued or simply change posture. Every one of these changes modifies the biomechanics of my body and shifts its center of mass. The brain never knows exactly the current mechanical state of the body. Instead, before every movement it generates an internal prediction. It estimates the body’s current dynamics and plans the movement accordingly. Then—and this is the fascinating part—it deliberately allows a small prediction error to occur.
That error is not a failure. It is information. By measuring the discrepancy between the predicted movement and the actual outcome, the brain continuously updates its internal model of the body. This happens constantly. Every time you brush your teeth, every time you reach for a cup of coffee, every time you pick up an object, your brain predicts, observes the error and recalibrates the system. In other words, we continuously relearn our own body.
So even errors are part of normal motor control?
Absolutely. Think about walking down the street. Occasionally you fail to lift your foot high enough, you slightly stumble and immediately take a corrective step.
We’ve all experienced this. What happened? Your brain immediately estimated the magnitude of the error, updated its prediction, and corrected the movement. The important point is that this error was internally generated. The nervous system was actively trying to perform the task, made a small mistake, detected it and learned from it. This is how motor learning occurs.
And this is where Functional Electrical Stimulation becomes particularly important?
Exactly. Functional Electrical Stimulation activates the patient’s own muscles. When the muscles contract, the patient attempts the movement, generates these internal prediction errors, and the brain begins correcting them. The cycle repeats over and over again. Prediction. Movement. Error. Correction. Learning. If the movement is produced only externally—if a robot simply moves the patient’s arm without engaging their own motor system—that same learning process does not occur in the same way. This was only a hypothesis initially. So, I decided to test it experimentally.
How did you test it?
We built a robotic system that does not move the patient’s arm. Instead, it presents objects within and just beyond the person’s reachable workspace: approximately 75% within reach and 25% outside their current capability.
The patient must actively attempt to reach, grasp and manipulate each object. If they fail, the robot brings it slightly closer and asks them to try again. In this way, each attempt generates prediction errors, and each correction contributes to motor learning. The robot does not replace movement; it creates the conditions under which the nervous system can learn.
In our pilot studies with patients with stroke and spinal cord injury, all participants increased their active range of motion against gravity. We also observed clinically meaningful functional improvements after approximately twenty hours of training. This supports the principle that effective rehabilitation is based on error-driven motor learning and that technology should facilitate this process rather than replace it.
How does functional electrical stimulation contribute to sensory recovery as well as motor activation?
This is a crucial and often overlooked issue. Muscle spindles provide the nervous system with information about muscle length, movement, and limb position and velocity.
After stroke or spinal cord injury, this system may be severely altered. In spasticity, muscles and associated sensory structures may remain excessively active; in flaccid paralysis, insufficient activation may produce very limited sensory feedback. In both cases, proprioceptive information reaching the brain is compromised.
When electrical stimulation activates a muscle, it also affects the sensory afferents associated with the muscle spindle. The resulting feedback may be imperfect, but it is still more informative than no feedback at all.
I sometimes compare it to badly singing a Michael Jackson song: the performance may be poor, but you can still recognize the song. Similarly, the brain may receive an imprecise sensory signal yet still recognize what the body is trying to do. With repeated activation and movement, the sensorimotor feedback loop can begin to re-engage.
Is there experimental evidence that this sensory feedback is necessary for recovery?
Animal experiments indicate that intact sensory afferent pathways are essential to training-induced improvement. When the afferent fibers supplying this proprioceptive system are disrupted, animals do not recover effectively through training. When the pathways remain intact and the animals undergo repeated training, improvement is observed.
These experiments support the broader principle that rehabilitation cannot be reduced to muscle contraction alone. Recovery depends on the interaction among motor commands, muscle activation and the sensory information returning to the spinal cord and brain.
Can functional electrical stimulation also have a role in addressing age-related functional decline?
I am not a geriatrician or an ageing specialist, so this interpretation should be treated cautiously. However, ageing is commonly associated with loss of muscle mass and strength, a process further aggravated by inactivity. As muscle volume decreases, the physical capacity to execute motor commands and respond to perturbations is reduced. This can affect balance, locomotion and everyday activities. Older adults may know what movement they want to perform but lack sufficient strength to carry it out effectively.
Functional electrical stimulation can help initiate muscle activation and support the preservation or development of muscle mass. It cannot recruit every muscle and should not replace exercise. Physical activity, including resistance exercise and calisthenics, remains essential, potentially combined with FES where appropriate. Maintaining muscle strength supports movement and balance, but also glucose metabolism, cardiovascular function and venous circulation. It also improves the ability to respond to unexpected disturbances and reduces the risk of losing balance.
Which technologies do you believe will have the greatest impact on neurorehabilitation over the next decade?
For upper-limb and hand rehabilitation, we already have a strong understanding of how functional electrical stimulation can be used. For the lower limbs, major challenges remain unresolved. Trunk control is particularly difficult because the trunk consists of many segments and degrees of freedom, and many of the core muscles are located deep within the body and are difficult to access through surface electrical stimulation.
Yet these muscles are essential for sitting, standing, balance and locomotion. This remains a significant technological challenge.
The most successful future technologies will be those that recognize that the brain, spinal cord and muscles are not separate systems. They form one distributed sensorimotor system. Whether the intervention involves robotics, virtual reality, FES or another technology, it must train the system as an integrated whole.
A technology that treats muscle simply as an actuator, independently of the brain and spinal cord, will be limited. A technology that engages motor intention, muscle activation, sensory feedback and the communication among these components has a much greater chance of producing meaningful recovery.
As director of a major multidisciplinary research institute, how do you see research centers evolving? What conditions are necessary for research to generate real clinical and societal impact?
We are facing an unprecedented demographic transformation, with rapid population ageing and a growing number of people living with disability and chronic conditions. Research institutes therefore have a major responsibility: governments will increasingly ask them for practical solutions to problems such as falls, impaired balance, dementia, mobility limitations and independent living.
Scientific publications remains important, but research risks becoming irrelevant if it does not improve healthcare, support older people or help health systems manage growing demand. Institutes must focus on challenges they can address within realistic timeframes and on interventions capable of producing clinically meaningful impact.
The priority is not simply to generate knowledge, but to translate it into technologies, clinical practices and services that solve real problems for large numbers of people. This requires multidisciplinary collaboration, a clear understanding of societal needs and an evaluation of research based not only on publications, but on the difference it makes in people’s lives.
