Rehabilitation and Health Care Robotics 53.4 Smart Prostheses and Orthoses 1241 tions of the nerve associated with different muscles in Invasive approaches involve implanting electrodes the elbow.wrist,and hand,and innervated three bundles on the surface of the brain (electrocorticogram)or into of the pectoralis muscle.After three months,the nerve the brain itself.Electrodes implanted inside the brain reinnervated the bundles so that the patient could cause can detect action potentials from single neurons.The the bundles to twitch by trying to bend his missing elbow, first demonstration of this technique in humans [53.137] for example.Surface EMG electrodes were placed over was in a person with end-stage amyotrophic lateral scle- the bundles.Then,when the user willed to open his hand, rosis (ALS),a disease that paralyzes muscles but leaves for example,a pectoralis muscle bundle contracted,and cognition intact.A cone electrode was implanted in the this contraction was detectable with the EMGelectrodes.motor cortex,along with growth factors that encour- The EMG signal was in turn used to control the hand aged growth of neurites,or branches of neurons,into the motor of the prosthetic arm.The net result was that the cone.Stable action potentials were recorded for several user could will his different(missing)anatomical joints months,and the patient was able to increase or decrease to move,and the corresponding joints on the robotic the firing rate of the action potentials recorded by the arm would move.He could simultaneously operate two electrode. joints,such as the elbow and the hand.The user be- Subsequent work in monkeys demonstrated that came able to do tasks that he was not able to do before recordings from multiple neurons(ranging from as few with his conventional myoelectric controlled arm,such as tens to hundreds)can be used in real time to de- as feeding himself,shaving,and throwing a ball.A sec- code the three-dimensional trajectory of the arm using ondary remarkable finding was that the sensory neurons straightforward signal processing(see review [53.138]). in the rerouted nerves reinnervated sensors,so that now The first human volunteer,a person with tetraplegia due when the person's chest is touched,the person perceives to spinal cord injury,has now been implanted with the art it as a touch to his missing limb.This sensory reinnerva- BrainGate electrode array,and has been able to control tion could possibly be made into an interface to provide the movement of a cursor on a computer screen [53.135]. tactile sensation from the artificial limb.These findings Other work in brain-machine interfaces has focused were recently confirmed in another person who received on detecting higher-level control signals,such as the targeted reinnervation [53.133].It has also recently been intent to move,preferences for different rewards,and shown that direct electrical stimulation of a residual pe- motivation to perform a task [53.139]. ripheral nerve can provide usable information regarding Given this progress,it appears that future control force to a person with an amputation [53.134]. systems for smart prosthetics and orthoses will have the option to rely on direct interfaces to the brain,which 53.4.3 Brain-Machine Interfaces should allow control of multiple joints through thought alone.The initial work on both targeted reinnervation There has also recently been progress in decoding and brain-machine interfaces has allowed three to four movement-related signals in real-time directly from the degrees of freedom of control in a naturalistic manner, brain(see the cover story and related articles in Nature, which is an advance over conventional prosthetic control Vol.442 [53.135)).The ability to decode an intended techniques. movement directly from brain activity could make it possible to control a prosthetic limb or orthotic device 53.4.4 Advances in Neural Stimulation directly by thought. Brain-computer interfaces can be divided into Unlike people who suffer an amputation,individuals non-invasive and invasive approaches.In noninva- who suffer from paralysis due to neurologic injury retain sive approaches,electrical activity is recorded from their limbs.Frustratingly,however,they often do not the surface of the skull using surface electrodes to achieve as good of control as people with artificial limbs. form an electroencephalogram (EEG)(for a review The situation is ironic because the actuator in the limb see [53.136]).Individuals can learn to control the am-muscle,is very sophisticated and can still be activated by plitude of the EEG signal as a function of time,or the stimulating the nerves that innervate the muscles.Thus. amplitude of specific frequency components of the EEG one viewpoint of the problem of restoring motion for signal,with a moderate amount of practice(several hours paralyzed people is that excellent hardware for solving to several days).The level of control is sufficient to op- the problem already exists (i.e.,the limbs themselves), erate a typing program on a computer,or to control the but that we must find ways to replace the control system movement of a cursor to multiple targets. to control it.Rehabilitation and Health Care Robotics 53.4 Smart Prostheses and Orthoses 1241 tions of the nerve associated with different muscles in the elbow, wrist, and hand, and innervated three bundles of the pectoralis muscle. After three months, the nerve reinnervated the bundles so that the patient could cause the bundles to twitch by trying to bend his missing elbow, for example. Surface EMG electrodes were placed over the bundles. Then, when the user willed to open his hand, for example, a pectoralis muscle bundle contracted, and this contraction was detectable with the EMGelectrodes. The EMG signal was in turn used to control the hand motor of the prosthetic arm. The net result was that the user could will his different (missing) anatomical joints to move, and the corresponding joints on the robotic arm would move. He could simultaneously operate two joints, such as the elbow and the hand. The user be￾came able to do tasks that he was not able to do before with his conventional myoelectric controlled arm, such as feeding himself, shaving, and throwing a ball. A sec￾ondary remarkable finding was that the sensory neurons in the rerouted nerves reinnervated sensors, so that now when the person’s chest is touched, the person perceives it as a touch to his missing limb. This sensory reinnerva￾tion could possibly be made into an interface to provide tactile sensation from the artificial limb. These findings were recently confirmed in another person who received targeted reinnervation [53.133]. It has also recently been shown that direct electrical stimulation of a residual pe￾ripheral nerve can provide usable information regarding force to a person with an amputation [53.134]. 53.4.3 Brain–Machine Interfaces There has also recently been progress in decoding movement-related signals in real-time directly from the brain (see the cover story and related articles in Nature, Vol. 442 [53.135]). The ability to decode an intended movement directly from brain activity could make it possible to control a prosthetic limb or orthotic device directly by thought. Brain–computer interfaces can be divided into non-invasive and invasive approaches. In noninva￾sive approaches, electrical activity is recorded from the surface of the skull using surface electrodes to form an electroencephalogram (EEG) (for a review see [53.136]). Individuals can learn to control the am￾plitude of the EEG signal as a function of time, or the amplitude of specific frequency components of the EEG signal, with a moderate amount of practice (several hours to several days). The level of control is sufficient to op￾erate a typing program on a computer, or to control the movement of a cursor to multiple targets. Invasive approaches involve implanting electrodes on the surface of the brain (electrocorticogram) or into the brain itself. Electrodes implanted inside the brain can detect action potentials from single neurons. The first demonstration of this technique in humans [53.137] was in a person with end-stage amyotrophic lateral scle￾rosis (ALS), a disease that paralyzes muscles but leaves cognition intact. A cone electrode was implanted in the motor cortex, along with growth factors that encour￾aged growth of neurites, or branches of neurons, into the cone. Stable action potentials were recorded for several months, and the patient was able to increase or decrease the firing rate of the action potentials recorded by the electrode. Subsequent work in monkeys demonstrated that recordings from multiple neurons (ranging from as few as tens to hundreds) can be used in real time to de￾code the three-dimensional trajectory of the arm using straightforward signal processing (see review [53.138]). The first human volunteer, a person with tetraplegia due to spinal cord injury, has now been implanted with the BrainGate electrode array, and has been able to control the movement of a cursor on a computer screen [53.135]. Other work in brain–machine interfaces has focused on detecting higher-level control signals, such as the intent to move, preferences for different rewards, and motivation to perform a task [53.139]. Given this progress, it appears that future control systems for smart prosthetics and orthoses will have the option to rely on direct interfaces to the brain, which should allow control of multiple joints through thought alone. The initial work on both targeted reinnervation and brain–machine interfaces has allowed three to four degrees of freedom of control in a naturalistic manner, which is an advance over conventional prosthetic control techniques. 53.4.4 Advances in Neural Stimulation Unlike people who suffer an amputation, individuals who suffer from paralysis due to neurologic injury retain their limbs. Frustratingly, however, they often do not achieve as good of control as people with artificial limbs. The situation is ironic because the actuator in the limb muscle, is very sophisticated and can still be activated by stimulating the nerves that innervate the muscles. Thus, one viewpoint of the problem of restoring motion for paralyzed people is that excellent hardware for solving the problem already exists (i. e., the limbs themselves), but that we must find ways to replace the control system to control it. Part F 53.4