These 'Bots Are Made for Walking
Walking seems so simple: Just put one foot in front of the other. Yet every step you take is a precarious act. When you walk, your body’s center of mass is rarely located over one of your feet. Each step forward is really a fall that you must catch by swinging your leg forward so that your foot lands on the floor just in the nick of time. The successful execution and timing of each step is aided by the way that the leg swings like a pendulum, but it still requires precise muscular coordination and excellent balance.
People who suffer a stroke, brain injury, or a partial spinal cord injury lose some of these essential abilities. Damage to the brain itself, or interference in the communication between the brain and the muscles, results in weakness or unwanted muscle activity, which limits the person’s ability to simultaneously propel and support the body during locomotion.
Fortunately, it turns out that the body has some built-in redundancy, at least in some species. Although the brain normally directs bodily motions, animal studies show that complex movements of the limbs are possible even after the spinal cord has been completely cut off from the brain. Such independence was first documented more than a century ago by British neurophysiologist Sir Charles Sherrington.
In the early 1900s Sherrington and others experimented on dogs and cats whose spinal cords had been severed (referred to as “spinal” animals). The studies revealed that the stepping motions of walking can be produced by spinal reflexes alone. The stepping reflexes discovered in those experiments were much more complicated and coordinated than the simple knee jerk that is elicited by tapping a rubber hammer on the tendon below the kneecap. Sherrington observed rhythmic stepping motions in which one limb of the animal was drawn toward the body while the limb on the opposite side of the body was pushed away—all without any master coordination by the brain.
Those findings in spinal cats and dogs suggested that control of locomotion in healthy, intact animals originates at least in part in the spinal cord rather than in the brain. But it took many decades before anyone found practical applications for that insight. In the 1970s researchers showed that if the body of a spinal cat were supported over a treadmill, the cat’s legs would step in time with the moving belt. When the same types of experiments were performed on spinal kittens also given a drug to enhance nerve signal transmission, the animals’ stepping motions would even adapt to changes in the speed of the treadmill.
More amazing, the stepping behavior of spinal cats improves with practice, implying that their spinal cords, not their brains, have done the learning. Daily training on the treadmill causes spinal cats to take longer steps, achieve more normal muscle excitation patterns, and support more of their own weight, although learning appears to be specific to the activity that is practiced: Cats that are trained to stand do poorly on locomotion tests.
The animal results inevitably raised human questions: Do people have the same neural circuitry that permits reflexive locomotion in cats? And if so, is recovery of walking ability possible for patients with injuries to the brain or spinal cord? Anecdotal evidence points to such “central pattern generators” for locomotion in humans. For instance, there have been reports of involuntary stepping motions in patients who suffered incomplete spinal cord injury. Also, newborn babies make alternating stepping motions with their legs, displaying timing patterns that roughly correspond to those seen years later in their mature walking.
Relearning to Walk
Encouraged by those insights, neuroscientists and physical therapists have attempted to adapt the training procedures for spinal animals into therapeutic regimens for patients who have an incomplete spinal cord injury or are recovering from a stroke. Those programs have largely taken the form of body weight-supported treadmill training, in which the patient’s torso is suspended above the treadmill using a harness while two therapists (one for each leg) manually move the legs through the repetitive motions of normal walking.
Treadmill training has been shown to improve strength, stability, and walking ability, but the technique has a significant drawback: It is highly labor-intensive. At least two therapists are needed to move the patient’s legs, and even a single 30-minute training session is fatiguing for the therapists’ arms. Something better was needed. And this time the answer came not from animal studies, but from the world of robotics.
Robotic arms have been common in industrial settings for decades, but recent reductions in the cost and size of sensors, actuators, and computers have led to a proliferation of applications for robots outside of factories. One exciting development is the robotic exoskeleton, a motorized, jointed scaffolding attached around the body to enhance strength and other abilities. Some experimental exoskeletons have been developed with funding from the military to improve a soldier’s ability to carry heavy loads or march over long distances, but exoskeletons have also been designed to assist the disabled, including those undergoing treadmill gait training in rehabilitation.
The most well known and well studied of these rehab devices is the Lokomat, developed by a Swiss team in the 1990s. It looks like a treadmill attached to a small crane, from which the patient is suspended in a harness. The patient’s legs are strapped into an adjustable linkage that has motors and sensors located at the hip and knee joints, as well as passive straps that pull up the patient’s toes to keep them from hitting the treadmill belt when the leg swings forward. The Lokomat exoskeleton’s motors do the work of moving the legs, taking over the job that otherwise would need to be done by human therapists, while the sensors measure the hip and knee joint angles. A computerized controller uses the sensor readings to calculate the forces and torques that the Lokomat applies to the patient’s body.
To assess the effectiveness of robotic rehab training, researchers have compared it with conventional therapy, which involves tasks such as strength and balance exercises, walking between parallel bars, and walking unassisted on a treadmill. Which of those exercises a patient performs depends on the therapist’s judgment of his or her progress. Several groups have looked at stroke patients who are randomly selected for either robotic or conventional therapy, but the results so far are inconclusive. Some studies suggest that walking improves more after robotic training, whereas others suggest the opposite, and still more report no discernible difference between the two approaches.
Other studies have compared the results of robotic gait training with those from treadmill training in which therapists manually move patients’ legs. Here again the results are mixed, but it seems that robotic training offers the greatest benefits to patients early in their rehabilitation programs, when they may not be able to walk at all without the assistance of a robot.
Strategies for Robot Therapy
Those preliminary evaluations probably understate the potential of robot-assisted therapy, perhaps significantly, because researchers do not yet have a clear idea of the best way for a robot to interact with a patient’s legs. Robots have made impressive inroads in industry because of their capacity to perform the same tasks over and over with high precision, and specialists in robot rehabilitation initially employed those machines the same way: Robots would guide patients as they moved their legs on the treadmill, with the exoskeleton providing consistent corrective assistance when its sensors detected the patient deviating from a predetermined normal gait pattern. The end result is that the patients’ legs would always move normally, with the patients doing what they could with their muscles, and the exoskeleton making up the difference with its motors.
That approach, called position control, is an excellent way to program a robot tasked with machining identical engine parts on an assembly line, but it may not be ideal for helping people relearn to walk. True, moving the patient through a normal walking motion would show him or her what such movements feel like, and it might even generate sensory signals in the patient’s own nerves that would be helpful in relearning gait. Robot-guided motions would be preferable to no motion at all for patients with a severe disability. On the other hand, position control may rob patients of the motivation to generate normal motions because the robot produces all the movement patterns for them.
David Reinkensmeyer of the University of California at Irvine calls that possible loss of motivation the “slacking hypothesis.” When a robot guides limb motions, the patient’s body may adapt automatically by minimizing energy expenditure. Such compensation presents a problem, Reinkensmeyer says, because the rehabilitation patient needs to expend effort to learn how to walk again.
If you have ever taught a child to ride a bike, you know that when a motor skill is first learned, there is considerable fluctuation in the movement. But does that variation represent deviation that the nervous system will attempt to eliminate, or does it actually aid in learning? Russian neurophysiologist Nikolai Bernstein theorized that movement variation allows the nervous system to practice solving the problem of planning motion, in much the same way that a fourth-grader develops computational skills by completing a worksheet of many different long division problems. “Repetition without repetition,” according to Bernstein, is the key to motor learning.
The challenge, then, is to find the best way for the robot to allow the patient to deviate from perfect repetition. To that end, several groups are now exploring an alternative to position control, called impedance control. Mechanical impedance is resistance to motion; the idea behind impedance control is that the robot would simulate a springlike resistance to deviation from the desired movement trajectory. The spring might be very soft or even nonexistent for small deviations, giving patients the opportunity to learn how to correct their own errors. Larger deviations would be met with greater resistance that would move the patient along the preferred path and perhaps prevent him or her from becoming frustrated early in the course of rehabilitation.
Robots that guide patient motions in this way may behave much like the way our own limbs respond when we encounter a disturbance, such as when someone jostles your arm while you reach for a cup of coffee. In a 2004 study, kinesiologists David Franklin and Theodore Milner of Simon Fraser University in Canada, along with Udell So and Mitsuo Kawato of the Advanced Telecommunications Research Institute in Kyoto, asked healthy people to perform reaching movements while gripping a handle attached to a robot arm that attempted to destabilize their motion. Test subjects were still able to make straight movements by modifying the springlike behavior of their own limbs: They stiffened their arms in the direction perpendicular to the path of their reach, but not in the movement direction.
A similar strategy underlies a variant of impedance control called path control. In this approach, the robot produces less resistance along the desired line of movement so that patients have more control over the timing of their motions. The legs are therefore free to move when the patient propels them in the right direction, but the robot applies a gentle correction when they stray too far from the prescribed path, creating virtual “walls” that prevent too much sideways deviation.
Exaggerating Your Faults
As promising as impedance control may be, that kind of therapy presumes that the patient has a nervous system capable of accurately sensing and responding to deviations from correct limb motions. Patients who have difficulty walking because of conditions such as incomplete spinal cord injury or stroke frequently also have sensory deficits that prevent them from making full use of neural feedback from their muscles, tendons, and joints.
Fortunately, there are ways for sensors to provide this missing feedback by translating the changing configuration of a robotic exoskeleton into an animation on a virtual reality (VR) display that the patient views on a video screen. Such electronic sensors can also produce cleaner signals of motion than the noisy biological sensors in our bodies.
In 2009 Anat Mirelman and Judith Deutsch of the University of Medicine and Dentistry of New Jersey and Paolo Bonato of Harvard Medical School put the theory to the test. They found that stroke patients undergoing robotic gait training were able to walk farther and faster when their therapy was enhanced with a VR display. The researchers attributed the improvement partly to VR showing the subjects how their limbs were moving, and partly to the display functioning like a video game that kept patients engaged during training.
If patients relearning how to walk benefit from knowing the true positions of their legs, some neuroscientists have reasoned, why stop at telling them the truth? Patients may be able to learn even better and faster through error augmentation, which sounds much better than “lying.” In this scheme, when the patient moves in a manner that deviates from what is desired, an exaggerated depiction of the deviation is presented to him or her on a screen. Like a teacher circling mistakes on a test with a red pen, robotic sensors’ output may be amplified to draw the patient’s attention to movement error and thus motivate stronger and faster corrections.
Curiously, it seems that patients learn more rapidly when their errors are moderately embellished, but there is no benefit in telling big whoppers to the nervous system. James Patton of the University of Illinois at Chicago explains that there seems to be a “sweet spot” for achieving the greatest benefit from amplifying errors. Early trials indicate that there has to be enough magnification to involve more of the nervous system in the learning process, according to Patton, but not so much that the patient begins to doubt that he’s really observing his own behavior.
Age of the Exoskeleton
Although the first Lokomat was introduced two decades ago, researchers are still just beginning to explore the ways robots can help injured patients walk again. If robotic exoskeletons do become a fixture in the gait rehabilitation clinic, the role of the physical therapist is likely to change. Early on, some therapists worried that the engineers’ goal was to replace them with robots, but it is now clear that robots have great potential for freeing therapists to do what they do best. Robots do not mind the drudgery of working with patients on repetitive motions, which allows the therapists to concentrate on helping patients master the more intricate and varied tasks of daily living.
As robotic gait therapy matures, it will need to become more versatile. Walking on a treadmill is an excellent first activity, but patients need to do more than put in long hours walking at a constant speed in a straight line. Adding turns, stops and starts, and obstacles such as curbs will make rehab training more practical and more useful. Ultimately my colleagues and I hope to see robots help patients regain the ability to walk freely on normal ground. Newer, more compact and battery-powered exoskeletons that can operate away from the treadmill and even outside the clinic should help meet this objective. Eventually robots may be used to train patients to perform more complex everyday activities, such as climbing stairs, negotiating ramps, and getting into and out of cars and bathtubs.
“Robotic therapy” may evoke images of a powerful automaton yanking on one’s limbs, but the reality will be a lot less intimidating. In 2040, barring unforeseen circumstances, I will be 72 years old. By then there will be 80 million other Americans age 65 or older, twice the number of older adults in the United States today. I hope that I won’t have any serious mobility problems to overcome, but it is inevitable that many of my fellow elderly will be dealing with the consequences of strokes, limb amputations, joint replacements, and falls. And it is very likely that robots will be helping.
The public has already grown accustomed to smartphones, smart appliances, and even smart cars. The robots we encounter in physical therapy tomorrow will not be very different from smart versions of the exercise machines we use today. We look to the heart rate monitor on an exercise bike to keep from slacking off or overtaxing our cardiovascular systems. Rehabilitation robots will be used in a similar fashion to get a customized workout for the nervous system. My expectation is that my generation will not only welcome robotic therapy assistants, we will demand them.
- Franklin, D. W., U. So, M. Kawato, and M. E. Milner. 2004. Impedance control balances stability with metabolically costly muscle activation. Journal of Neurophysiology 92:3097–3105.
- Krebs, H. I., and N. Hogan. 2012. Robotic therapy: The tipping point. American Journal of Physical Medicine & Rehabilitation 91:S290–S297.
- Mirelman, A., P. Bonato, and J. E. Deutsch. 2009. Effects of training with a robot–virtual reality system compared with a robot alone on the gait of individuals after stroke. Stroke 40:169–174.
- Reinkensmeyer, D. J., and J. L. Patton. 2009. Can robots help the learning of skilled actions? Exercise and Sport Sciences Reviews 37:43–51.
- Rossignol, S., et al. 1996. Locomotor capacities after complete and partial lesions of the spinal cord. Acta Neurobiologiae Experimentalis 56:449–463.