Monkey Arm Research Articles

Recurrent neural networks controlling musculoskeletal models predict motor cortex activity during novel limb movements.

Goal-driven networks trained to perform a task analogous to that performed by biological neural populations are being increasingly utilized as insightful computational models of motor control. The resulting dynamics of the trained networks are then analyzed to uncover the neural strategies employed by the motor cortex to produce movements. However, these networks do not take into account the role of sensory feedback in producing movement, nor do they consider the complex biophysical underpinnings of the underlying musculoskeletal system. Moreover, these models can not be used in context of predictive neuromechanical simulations for hypothesis generation and prediction of neural strategies during novel movements. In this research, we adapt state-of-the-art deep reinforcement learning (DRL) algorithms to train a controller to drive a developed anatomically accurate monkey arm model to track experimentally recorded kinematics. We validate that the trained controller mimics biologically observed neural strategies to produce movement. The trained controller generalizes well to unobserved conditions as well as to perturbation analyses. The recorded firing rates of motor cortex neurons can be predicted from the controller activity with high accuracy even on unseen conditions. Finally, we validate that the trained controller outperforms existing goal-driven and representational models of motor cortex in single neuron decoding accuracy, thus showing the utility of the complex underpinnings of anatomically accurate models in shaping motor cortex neural activity during limb movements. The learned controller can be used for hypothesis generation and prediction of neural strategies during novel movements and unobserved conditions.

Spatiotemporal Maps of Proprioceptive Inputs to the Cervical Spinal Cord During Three-Dimensional Reaching and Grasping

Proprioceptive feedback is a critical component of voluntary movement planning and execution. Neuroprosthetic technologies aiming at restoring movement must interact with it to restore accurate motor control. Optimization and design of such technologies depends on the availability of quantitative insights into the neural dynamics of proprioceptive afferents during functional movements. However, recording proprioceptive neural activity during unconstrained movements in clinically relevant animal models presents formidable challenges. In this work, we developed a computational framework to estimate the spatiotemporal patterns of proprioceptive inputs to the cervical spinal cord during three-dimensional arm movements in monkeys. We extended a biomechanical model of the monkey arm with ex-vivo measurements, and combined it with models of mammalian group-Ia, Ib and II afferent fibers. We then used experimental recordings of arm kinematics and muscle activity of two monkeys performing a reaching and grasping task to estimate muscle stretches and forces with computational biomechanics. Finally, we projected the simulated proprioceptive firing rates onto the cervical spinal roots, thus obtaining spatiotemporal maps of spinal proprioceptive inputs during voluntary movements. Estimated maps show complex and markedly distinct patterns of neural activity for each of the fiber populations spanning the spinal cord rostro-caudally. Our results indicate that reproducing the proprioceptive information flow to the cervical spinal cord requires complex spatio-temporal modulation of each spinal root. Our model can support the design of neuroprosthetic technologies as well as in-silico investigations of the primate sensorimotor system.

Predicting voluntary movements from motor cortical activity with neuromorphic hardware

Neurons in the mammalian motor cortex encode physical parameters of voluntary movements during planning and execution of a motor task. Brain–machine interfaces can decode limb movements from the activity of these neurons in real time. One future goal is to control prosthetic devices in severely paralyzed patients or to restore communication if the ability to speak or make gestures is lost. Here, we implemented a spiking neural network that decodes movement intentions from the activity of individual neurons recorded in the motor cortex of a monkey. The network runs on neuromorphic hardware and performs its computations in a purely spike-based fashion. It incorporates an insect-brain-inspired, three-layer architecture with 176 neurons. Cortical signals are filtered using lateral inhibition, and the network is trained in a supervised fashion to predict two opposing directions of the monkey's arm reaching movement before the movement is carried out. Our network operates on the actual spikes that have been emitted by motor cortical neurons, without the need to construct intermediate nonspiking representations. Using a pseudo-population of 12 manually selected neurons, it reliably predicts the movement direction with an accuracy of 89.3% on data not encountered during model training, after only 100 training trials. Our results provide a proof of concept for the first-time use of a neuromorphic device for decoding movement intentions.

DEVISING A ROBOTIC ARM MANIPULANDUM FOR NORMAL AND ALTERED REACHING MOVEMENTS TO INVESTIGATE BRAIN MECHANISMS OF MOTOR CONTROL

The objective of this work was to develop a manipulandum system capable of measuring a monkey's arm movements and applying extrinsic forces to the hand. The manipulandum (RANARM) was designed as a five-bar linkage mechanism in order to realize sufficient rigidity and efficient force transmission. In addition, RANARM's control enables it to cancel its own dynamics by calculating the inverse dynamics. We evaluated the performance of RANARM from the mechanism, control, and application points of view. From the mechanism point of view, RANARM was able to provide a sufficiently large workspace for experiments, where it exhibited high manipulability and the capacity to apply force perturbations of at least 12 N under static conditions, and 5.3 N under dynamic conditions. From the control point of view, we estimated the uncompensated dynamics of RANARM to be negligibly small. Finally, we evaluated the performance of a monkey operating RANARM, and observed that the hand speed profiles followed a natural bell shape, while the maximal speeds were high (i.e., 15-cm reaching within 500 ms). In conclusion, RANARM allows a monkey to execute rapid movement and it is an efficient apparatus for investigating brain mechanisms for generating voluntary movements.

Intrafascicular stimulation of monkey arm nerves evokes coordinated grasp and sensory responses

High-count microelectrode arrays implanted in peripheral nerves could restore motor function after spinal cord injury or sensory function after limb loss. In this study, we implanted Utah Slanted Electrode Arrays (USEAs) intrafascicularly at the elbow or shoulder in arm nerves of rhesus monkeys (n = 4) under isoflurane anesthesia. Input-output curves indicated that pulse-width-modulated single-electrode stimulation in each arm nerve could recruit single muscles with little or no recruitment of other muscles. Stimulus trains evoked specific, natural, hand movements, which could be combined via multielectrode stimulation to elicit coordinated power or pinch grasp. Stimulation also elicited short-latency evoked potentials (EPs) in primary somatosensory cortex, which might be used to provide sensory feedback from a prosthetic limb. These results demonstrate a high-resolution, high-channel-count interface to the peripheral nervous system for restoring hand function after neural injury or disruption or for examining nerve structure.

Visual feedback effects as a coordination of joint stiffness in monkey's arm reaching

Visual feedback effects as a coordination of joint stiffness in monkey's arm reaching

Incorporating Feedback from Multiple Sensory Modalities Enhances Brain–Machine Interface Control

The brain typically uses a rich supply of feedback from multiple sensory modalities to control movement in healthy individuals. In many individuals, these afferent pathways, as well as their efferent counterparts, are compromised by disease or injury resulting in significant impairments and reduced quality of life. Brain-machine interfaces (BMIs) offer the promise of recovered functionality to these individuals by allowing them to control a device using their thoughts. Most current BMI implementations use visual feedback for closed-loop control; however, it has been suggested that the inclusion of additional feedback modalities may lead to improvements in control. We demonstrate for the first time that kinesthetic feedback can be used together with vision to significantly improve control of a cursor driven by neural activity of the primary motor cortex (MI). Using an exoskeletal robot, the monkey's arm was moved to passively follow a cortically controlled visual cursor, thereby providing the monkey with kinesthetic information about the motion of the cursor. When visual and proprioceptive feedback were congruent, both the time to successfully reach a target decreased and the cursor paths became straighter, compared with incongruent feedback conditions. This enhanced performance was accompanied by a significant increase in the amount of movement-related information contained in the spiking activity of neurons in MI. These findings suggest that BMI control can be significantly improved in paralyzed patients with residual kinesthetic sense and provide the groundwork for augmenting cortically controlled BMIs with multiple forms of natural or surrogate sensory feedback.

Relationship between Unconstrained Arm Movements and Single-Neuron Firing in the Macaque Motor Cortex

The activity of single neurons in the monkey motor cortex was studied during semi-naturalistic, unstructured arm movements made spontaneously by the monkey and measured with a high resolution three-dimensional tracking system. We asked how much of the total neuronal variance could be explained by various models of neuronal tuning to movement. On average, tuning to the speed of the hand accounted for 1% of the total variance in neuronal activity, tuning to the direction of the hand in space accounted for 8%, a more complex model of direction tuning, in which the preferred direction of the neuron rotated with the starting position of the arm, accounted for 13%, tuning to the final position of the hand in Cartesian space accounted for 22%, and tuning to the final multijoint posture of the arm accounted for 36%. One interpretation is that motor cortex neurons are significantly tuned to many control parameters important in the animal's repertoire, but that different control parameters are represented in different proportion, perhaps reflecting their prominence in everyday action. The final posture of a movement is an especially prominent control parameter although not the only one. A common mode of action of the monkey arm is to maintain a relatively stable overall posture while making local adjustments in direction during performance of a task. One speculation is that neurons in motor cortex reflect this pattern in which direction tuning predominates in local regions of space and postural tuning predominates over the larger workspace.

Somatosensory-Motor Neuronal Activity in the Superior Colliculus of the Primate

The superior colliculus (SC) in primates plays an important role in orienting gaze and arms toward novel stimuli. Here we ask whether neurons in the intermediate and deep layers of the SC are also involved in the interaction with objects. In two trained monkeys we found a large number of SC units that were specifically activated when the monkeys contacted and pushed a target that had been reached with either hand. These neurons, however, were silent when the monkeys simply looked at or reached for the target but did not touch it. The activity related to interacting with objects was spatially tuned and increased with push strength. Neurons in the SC with this type of activity may be involved in a somatosensory-motor feedback loop that monitors the force of the active muscles together with the spatial position of the limb required for proper interaction with an object.

Afferent Encoding of Central Oscillations in the Monkey Arm

We have investigated whether peripheral afferent fibers could encode the central oscillations that are commonly seen in the primate motor system. We analyzed 52 single afferent recordings from the C8/T1 dorsal root ganglia of two monkeys performing an isometric wrist flexion-extension task. Coherence and directed coherence were calculated between the afferent spikes and forearm EMG. Seven of 52 cells were identified as Group Ia afferents by the production of narrow postspike facilitation in spike-triggered averages of rectified EMG. These identified afferents showed significant coherence, and directed coherence, with EMG over a wide frequency range. By contrast, coherence was weak for a population that showed little directional preference for flexion or extension movements during task performance, and probably contained mainly cutaneous afferents. Oscillations are known to appear in muscle activity; their presence in afferent firing as well implies that central oscillations pass around a peripheral feedback loop and may be involved in sensorimotor integration.

Prediction of arm trajectory from a small number of neuron activities in the primary motor cortex

Monkey arm movement was reconstructed from neuron activities recorded in the primary motor cortex (Ml). We recorded single neuron activities from a monkey's Ml, while the animal performed an arm reaching task. We also recorded electromyographic (EMG) activity and movement trajectories during the task. First, we reconstructed the EMG signals from the neuron activities. The EMG signals were reliably reconstructed with a linear summation of the neuron activities. Next, we reconstructed joint angles from the reconstructed EMG signals with an artificial neural network model. The reconstructed trajectories of the hand position and elbow position showed good correlation with the actual arm movement. This model appropriately reflected the anatomical characteristics.

Target-, limb-, and context-dependent neural activity in the cingulate and supplementary motor areas of the monkey.

Very little is known about the role of the cingulate motor area (CMA) in visually guided reaching compared to other cortical motor areas. To investigate the hierarchical role of the caudal CMA (CMAc) during reaching we recorded the activity of neurons in CMAc in comparison to the supplementary motor area proper (SMA) while a monkey performed an instructed delay task that required it to position a cursor over visual targets on a computer screen using two-dimensional (2D) joystick movements. The direction of the monkey's arm movement was dissociated from the direction of the visual target by periodically reversing the relationship between the direction of movement of the joystick and that of the cursor. Neurons that responded maximally with a particular limb movement direction regardless of target location were classified as limb-dependent, whereas neurons that responded maximally to a particular target direction regardless of the direction of limb movement were classified as target-dependent. Neurons whose activity was directional in one of the two visuomotor mapping conditions and non-directional or inactive in the other were categorized as context-dependent. Limb-dependent activity was observed more frequently than target-dependent activity in both CMAc and SMA proper during both the delay period (preparatory activity; CMAc, 17%; SMA, 31%) and during movement execution (CMAc, 49%, SMA, 48%). A modest percentage of neurons with preparatory activity were target-dependent in both CMAc (11%) and SMA proper (8%) and a similar percentage of neurons in both areas demonstrated target-dependent, movement activity (CMAc, 8%; SMA, 10%). The surprising finding was that a very large percentage of neurons in both areas displayed context-dependent activity either during the preparatory (CMAc, 72%; SMA, 61%) or movement (CMAc, 43%, SMA 42%) epochs of the task. These results show that neural activity in both CMAc and SMA can directly represent movement direction in either limb-centered or target-centered coordinates. The presence of target-dependent activity in CMAc, as well as SMA, suggests that both are involved in the transformation of visual target information into appropriate motor commands. Target-dependent activity has been found in the putamen, SMA, CMAc, dorsal and ventral premotor cortex, as well as primary motor cortex. This indicates that the visuomotor transformations required for visually guided reaching are carried out by a distributed network of interconnected motor areas. The large proportion of neurons with context-dependent activity suggests, however, that while both CMAc and SMA may play a role in the visuomotor transformation of target information into movement parameters, their activity is not solely coding parameters of movement, since their involvement in this process is highly condition-dependent.

Effects of muscimol inactivation of the cerebellar nuclei on precision grip.

A single monkey was trained to perform a grasp, lift, and hold task in which a stationary hand- held object was sometimes subjected to brief, predictable force-pulse perturbations. The displacement, grip, and lifting forces were measured as well the three-dimensional forces and torques to quantify specific motor deficits after reversible inactivation of the cerebellar nuclei. A prior single-cell recording study in the same monkey provided the stereotaxic coordinates used to guide intranuclear injections of muscimol. In total, 34 penetrations were performed at 28 different loci throughout the cerebellar nuclei. On each penetration, two 1.0-microl injections of 5 microg/microl muscimol, were made 1.0 mm apart either within the nuclei or in the white matter just lateral or posterior to the dentate nucleus. Injections in the region corresponding to the anterior interpositus nucleus produced pronounced dynamic tremor and dysmetric movements of the ipsilateral arm when the animal performed unrestrained reaching and grasping movements. In contrast, no relatively short-latency (15-20 min.) deficits were observed after injection in the dentate nucleus, although some effects were observed after several hours. When tested in a primate chair with the forearm supported and restrained at the wrist and elbow, the monkey performed the lift and hold task without tremor or dysmetria. However, with the restraint removed, the forces and torques applied to the manipulandum were poorly controlled and erratic. The monkey's arm was ataxic and a 5-Hz intention tremor was clearly visible. In addition, the animal was generally unable to compensate for the predictable perturbations and the anticipatory grip force increases were absent. However, overall the results suggest that reversible cerebellar nuclear inactivation with muscimol has little effect on isolated distal movements of the wrist and fingers.

Coding the location of the arm by sight.

Area 5 in the parietal lobe of the primate brain is thought to be involved in monitoring the posture and movement of the body. In this study, neurons in monkey area 5 were found to encode the position of the monkey's arm while it was covered from view. The same neurons also responded to the position of a visible, realistic false arm. The neurons were not sensitive to the sight of unrealistic substitutes for the arm and were able to distinguish a right from a left arm. These neurons appear to combine visual and somatosensory signals in order to monitor the configuration of the limbs. They could form the basis of the complex body schema that we constantly use to adjust posture and guide movement.

Minimum Muscle-Tension Change Trajectories Predicted by Using a 17-Muscle Model of the Monkey's Arm.

Four computational problems to be solved for visually guided reaching movements, hand path, and trajectory formations, coordinate transformation, and calculations of muscle tensions are ill-posed in redundant biological control systems. These problems are ill-posed in the sense that there exist an infinite number of possible solutions. In this article, it is shown that the nervous system can solve those problems simultaneously by imposing a single global constraint: finding the smoothest muscle- tension trajectory that satisfies the desired final hand position, velocity, and acceleration. Horizontal trajectories were simulated by using a l7-muscle model of the monkey's arm as the controlled object. The simulations predicted gently curved hand paths for lateral hand movements and for movements from the side of the body to the front, and a roughly straight hand path for anterioposterior movements. The tangential hand velocities were roughly bell shaped. The simulated results were in agreement with the actual biological movements.

2806 A motor task which requires motor learning in monkey arm movements

2806 A motor task which requires motor learning in monkey arm movements

Equilibrium point control of a monkey arm simulator by a fast learning tree structured artificial neural network.

A planar 17 muscle model of the monkey's arm based on realistic biomechanical measurements was simulated on a Symbolics Lisp Machine. The simulator implements the equilibrium point hypothesis for the control of arm movements. Given initial and final desired positions, it generates a minimum-jerk desired trajectory of the hand and uses the backdriving algorithm to determine an appropriate sequence of motor commands to the muscles (Flash 1987; Mussa-Ivaldi et al. 1991; Dornay 1991b). These motor commands specify a temporal sequence of stable (attractive) equilibrium positions which lead to the desired hand movement. A strong disadvantage of the simulator is that it has no memory of previous computations. Determining the desired trajectory using the minimum-jerk model is instantaneous, but the laborious backdriving algorithm is slow, and can take up to one hour for some trajectories. The complexity of the required computations makes it a poor model for biological motor control. We propose a computationally simpler and more biologically plausible method for control which achieves the benefits of the backdriving algorithm. A fast learning, tree-structured network (Sanger 1991c) was trained to remember the knowledge obtained by the backdriving algorithm. The neural network learned the nonlinear mapping from a 2-dimensional cartesian planar hand position (x,y) to a 17-dimensional motor command space (u1, . . ., u17). Learning 20 training trajectories, each composed of 26 sample points [[x,y], [u1, . . ., u17] took only 20 min on a Sun-4 Sparc workstation. After the learning stage, new, untrained test trajectories as well as the original trajectories of the hand were given to the neural network as input. The network calculated the required motor commands for these movements. The resulting movements were close to the desired ones for both the training and test cases.

Spinal stretch reflex and cortical evoked potential amplitudes versus muscle stretch amplitude in the monkey arm

While investigating operant conditioning of the primate spinal stretch reflex (SSR), we studied SSR amplitude and cortical somatosensory evoked potential (SEP) amplitude as stretch amplitude changed in the monkey arm. Initial muscle length and background EMG activity remained constant. With change in stretch amplitude (and proportional change in stretch velocity and acceleration), changes in SSR and SEP amplitudes were respectively 0.75 and 0.66 as great. The lesser change in SSR amplitude may reflect saturation of Ia afferents, while that in SEP amplitude may also reflect participation of other peripheral receptors.

Responses to heat of C-fiber nociceptors in monkey are altered by injury in the receptive field but not by adjacent injury

We sought to determine the effects of a cut injury on the thermal responsiveness of C-fiber nociceptors sensitive to heat and mechanical stimuli (CMHs). Teased fiber techniques were used to record from single CMHs that innervated the hairy skin of the monkey arm. Responses to heat stimuli ranging from 41 to 49°C were compared before and after injury. In 11 CMHs, the injury was applied 4 mm peripheral to the edge of the receptive field. The response to the heat sequence was not significantly altered by this adjacent injury. In 16 CMHs, a cut was applied directly to the receptive field. This direct injury led to a significant increase in response to the sequence of heat stimuli (i.e., sensitization). It is concluded that spreading sensitization of C-fiber nociceptors to a cut injury does not occur in monkey.

Operant conditioning of specific patterns of neural and muscular activity.

In awake monkeys we recorded activity of single "motor" cortex cells, four contralateral arm muscles, and elbow position, while operantly reinforcing several patterns of motor activity. With the monkey's arm held semiprone in a cast hinged at the elbow, we reinforced active elbow movements and tested cell responses to passive elbow movements. With the cast immobilized we reinforced isometric contraction of each of the four muscles in isolation, and bursts of cortical cell activity with and without simultaneous suppression of muscle activity. Correlations between a precentral cell and specific arm muscles consistently appeared under several behavioral conditions, but could be dissociated by reinforcing cell activity and muscle suppression.