David Reinkensmeyer, University of California at Irvine
12 August 2013
Engineers and clinicians currently design robotic control algorithms for promoting neurorehabilitation in an ad-hoc manner because we still don't fully understand the fundamental structure of human motor control, neuroplasticity, and motor learning. If we better understood this structure, we could in principle mathematically derive the control algorithms that would help robotic devices help patients to optimize motor recovery. In this paper, Moreno et al. take a step forward in identifying the structure of control during walking in the most widely used robotic gait training device, the Lokomat.
The Lokomat, like the MIT-MANUS, is ingenious because it allows severely impaired individuals to recreate a basic appearance and feel of the target movement - stepping for the Lokomat and reaching for the MIT-MANUS - with a relatively simple mechanical design. There is now ample evidence that patients with stroke and spinal cord injury benefit from training with the Lokomat, although it is still unclear for what applications the Lokomat is better than other forms of gait training. However, the device has been criticized because it does not allow a fully naturalistic pattern of walking, blocking key motions such as the pelvic rotations and translations used for balance. In addition, initial control algorithms for the device that used relatively rigid forms of assistance promoted slacking by the users, possibly diminishing the benefits these users gained from training with the device.
Moreno et al. examine how walking in the Lokomat affects motor modules, which have previously been hypothesized to be fixed combinations of muscles that the locomotor system activates with time-varying patterns to create walking. As has been done previously, they identified the motor modules by applying non-negative matrix factorization to EMG recordings of key leg muscles from non-impaired subjects. They found that four modules were sufficient to explain much of the muscle activity of walking in the robot. Remarkably, as they varied the amount of assistance the robot provided as well as the walking speed in the device, the basic four modules identified with this mathematical technique remained nearly identical, and comparable to treadmill walking outside the robot, although the activation patterns of those modules changed to some degree (see their Figure 4). This result reaffirms the concept that the motor system may use a concise set of only four fixed combinations of muscle activations during walking. It further suggests that assisted walking in a somewhat restrictive robotic device does not alter the basic structure of locomotor control, viewed from the motor module viewpoint.
While robotic guidance does not appear to cause a wholesale change of control structure, the results of Moreno et al. suggest that robotic guidance alters how those building blocks are combined. A key question then becomes: how does one choose the controller and training conditions such that the way the motor system activates the modules during training maximizes recovery? The answer to this question will require studies of how guidance and other forms of robotic training affect learning and recovery, but this work supports the ongoing use of the mathematically-rich framework of motor modules for those studies.
Competing interests
David Reinkensmeyer is a co-inventor of a device licensed and sold by Hocoma A.G. and is a co-founder of Flint Rehabilitation Devices. The terms of these arrangements have been reviewed and approved by the University of California, Irvine, in accordance with its conflict of interest policies.
Comment on Moreno et al.
12 August 2013
Engineers and clinicians currently design robotic control algorithms for promoting neurorehabilitation in an ad-hoc manner because we still don't fully understand the fundamental structure of human motor control, neuroplasticity, and motor learning. If we better understood this structure, we could in principle mathematically derive the control algorithms that would help robotic devices help patients to optimize motor recovery. In this paper, Moreno et al. take a step forward in identifying the structure of control during walking in the most widely used robotic gait training device, the Lokomat.
The Lokomat, like the MIT-MANUS, is ingenious because it allows severely impaired individuals to recreate a basic appearance and feel of the target movement - stepping for the Lokomat and reaching for the MIT-MANUS - with a relatively simple mechanical design. There is now ample evidence that patients with stroke and spinal cord injury benefit from training with the Lokomat, although it is still unclear for what applications the Lokomat is better than other forms of gait training. However, the device has been criticized because it does not allow a fully naturalistic pattern of walking, blocking key motions such as the pelvic rotations and translations used for balance. In addition, initial control algorithms for the device that used relatively rigid forms of assistance promoted slacking by the users, possibly diminishing the benefits these users gained from training with the device.
Moreno et al. examine how walking in the Lokomat affects motor modules, which have previously been hypothesized to be fixed combinations of muscles that the locomotor system activates with time-varying patterns to create walking. As has been done previously, they identified the motor modules by applying non-negative matrix factorization to EMG recordings of key leg muscles from non-impaired subjects. They found that four modules were sufficient to explain much of the muscle activity of walking in the robot. Remarkably, as they varied the amount of assistance the robot provided as well as the walking speed in the device, the basic four modules identified with this mathematical technique remained nearly identical, and comparable to treadmill walking outside the robot, although the activation patterns of those modules changed to some degree (see their Figure 4). This result reaffirms the concept that the motor system may use a concise set of only four fixed combinations of muscle activations during walking. It further suggests that assisted walking in a somewhat restrictive robotic device does not alter the basic structure of locomotor control, viewed from the motor module viewpoint.
While robotic guidance does not appear to cause a wholesale change of control structure, the results of Moreno et al. suggest that robotic guidance alters how those building blocks are combined. A key question then becomes: how does one choose the controller and training conditions such that the way the motor system activates the modules during training maximizes recovery? The answer to this question will require studies of how guidance and other forms of robotic training affect learning and recovery, but this work supports the ongoing use of the mathematically-rich framework of motor modules for those studies.
Competing interests
David Reinkensmeyer is a co-inventor of a device licensed and sold by Hocoma A.G. and is a co-founder of Flint Rehabilitation Devices. The terms of these arrangements have been reviewed and approved by the University of California, Irvine, in accordance with its conflict of interest policies.