Internal Forward Model Research Articles

Emergent Synergistic Grasp-Like Behavior in a Visuomotor Joint Action Task: Evidence for Internal Forward Models as Building Blocks of Human Interactions.

Central to the mechanistic understanding of the human mind is to clarify how cognitive functions arise from simpler sensory and motor functions. A longstanding assumption is that forward models used by sensorimotor control to anticipate actions also serve to incorporate other people’s actions and intentions, and give rise to sensorimotor interactions between people, and even abstract forms of interactions. That is, forward models could aid core aspects of human social cognition. To test whether forward models can be used to coordinate interactions, here we measured the movements of pairs of participants in a novel joint action task. For the task they collaborated to lift an object, each of them using fingers of one hand to push against the object from opposite sides, just like a single person would use two hands to grasp the object bimanually. Perturbations of the object were applied randomly as they are known to impact grasp-specific movement components in common grasping tasks. We found that co-actors quickly learned to make grasp-like movements with grasp components that showed coordination on average based on action observation of peak deviation and velocity of their partner’s trajectories. Our data suggest that co-actors adopted pre-existing bimanual grasp programs for their own body to use forward models of their partner’s effectors. This is consistent with the long-held assumption that human higher-order cognitive functions may take advantage of sensorimotor forward models to plan social behavior.New and Noteworthy: Taking an approach of sensorimotor neuroscience, our work provides evidence for a long-held belief that the coordination of physical as well as abstract interactions between people originates from certain sensorimotor control processes that form mental representations of people’s bodies and actions, called forward models. With a new joint action paradigm and several new analysis approaches we show that, indeed, people coordinate each other’s interactions based on forward models and mutual action observation.

Cerebellum, Predictions and Errors.

Making predictions and validating the predictions against actual sensory information is thought to be one of the most fundamental functions of the nervous system. A growing body of evidence shows that the neural mechanisms controlling behavior, both in motor and non-motor domains, rely on prediction errors, the discrepancy between predicted and actual information. The cerebellum has been viewed as a key component of the motor system providing predictions about upcoming movements and receiving feedback about motor errors. Consequentially, studies of cerebellar function have focused on the motor domain with less consideration for the wider context in which movements are generated. However, motor learning experiments show that cognition makes important contributions to motor adaptation that involves the cerebellum. One of the more successful theoretical frameworks for understanding motor control and cerebellar function is the forward internal model which states that the cerebellum predicts the sensory consequences of the motor commands and is involved in computing sensory prediction errors by comparing the predictions to the sensory feedback. The forward internal model was applied and tested mainly for effector movements, raising the question whether cerebellar encoding of behavior reflects task performance measures associated with cognitive involvement. Electrophysiological studies based on pseudo-random tracking in monkeys show that the discharge of Purkinje cell, the sole output neurons of the cerebellar cortex, encodes predictive and feedback signals not only of the effector kinematics but also of task performance. The implications are that the cerebellum implements both effector and task performance forward models and the latter are consistent with the cognitive contributions observed during motor learning. The implications of these findings include insights into recent psychophysical observations on moving with reduced feedback and motor learning. The findings also support the cerebellum’s place in hierarchical generative models that work in concert to refine predictions about behavior and the world. Therefore, cerebellar representations bridge motor and non-motor domains and provide a better understanding of cerebellar function within the functional architecture of the brain.

Neural Evidence of the Cerebellum as a State Predictor

We here provide neural evidence that the cerebellar circuit can predict future inputs from present outputs, a hallmark of an internal forward model. Recent computational studies hypothesize that the cerebellum performs state prediction known as a forward model. To test the forward-model hypothesis, we analyzed activities of 94 mossy fibers (inputs to the cerebellar cortex), 83 Purkinje cells (output from the cerebellar cortex to dentate nucleus), and 73 dentate nucleus cells (cerebellar output) in the cerebro-cerebellum, all recorded from a monkey performing step-tracking movements of the right wrist. We found that the firing rates of one population could be reconstructed as a weighted linear sum of those of preceding populations. We then went on to investigate if the current outputs of the cerebellum (dentate cells) could predict the future inputs of the cerebellum (mossy fibers). The firing rates of mossy fibers at time t + t1 could be well reconstructed from as a weighted sum of firing rates of dentate cells at time t, thereby proving that the dentate activities contained predictive information about the future inputs. The average goodness-of-fit (R2) decreased moderately from 0.89 to 0.86 when t1 was increased from 20 to 100 ms, hence indicating that the prediction is able to compensate the latency of sensory feedback. The linear equations derived from the firing rates resembled those of a predictor known as Kalman filter composed of prediction and filtering steps. In summary, our analysis of cerebellar activities supports the forward-model hypothesis of the cerebellum.

Predictive coding and adaptive behavior in patients with genetically determined cerebellar ataxia--A neurophysiology study.

Genetically determined cerebellar ataxias (CA) are a heterogeneous group of disorders with progressive decline of cerebellar functions. The cerebellum influences internal forward models that play a role in cognitive control, but whether these processes are dysfunctional in CA is unclear. Here, we examined sensory predictive coding processes and response adaptation in CA and healthy controls (HC) using behavioral tests with concomitant EEG recordings. N = 23 patients and N = 29 age- and sex-matched HC were studied. Sensory prediction coding was tested with an auditory distraction paradigm and error-related behavioral adaptation with a visual flanker task. As neurophysiological markers we studied different event-related potentials: the P3a for orientation of attention; the N2 and the error-related negativity (ERN) for cognitive adaptation processes/consequences of response errors; error-related positivity (Pe) for error-awareness; the mismatch negativity (MMN) for sensory predictive coding; and reorientation negativity (RON) for reorientation after unexpected events. Overall reaction times were slower in patients compared to HC, but error rates did not differ. Both in patients and HC, P3a amplitudes were larger in distraction trials, but the P3a amplitude was smaller in patients compared to HC. The MMN as well as behavioral and EEG-correlates of response adaptation (ERN/N2) did not differ between groups, while the Pe was attenuated in patients. During sensory predictive coding, RON amplitudes were significantly larger in HC compared to patients. In HC, but not in patients, RON amplitudes were also larger in deviant compared to frequent trials. Processes generating internal forward models are largely intact in genetically determined CA, whereas updating of mental models and error awareness are disturbed in these patients.

Voice-selective prediction alterations in nonclinical voice hearers

Auditory verbal hallucinations (AVH) are a cardinal symptom of psychosis but also occur in 6–13% of the general population. Voice perception is thought to engage an internal forward model that generates predictions, preparing the auditory cortex for upcoming sensory feedback. Impaired processing of sensory feedback in vocalization seems to underlie the experience of AVH in psychosis, but whether this is the case in nonclinical voice hearers remains unclear. The current study used electroencephalography (EEG) to investigate whether and how hallucination predisposition (HP) modulates the internal forward model in response to self-initiated tones and self-voices. Participants varying in HP (based on the Launay-Slade Hallucination Scale) listened to self-generated and externally generated tones or self-voices. HP did not affect responses to self vs. externally generated tones. However, HP altered the processing of the self-generated voice: increased HP was associated with increased pre-stimulus alpha power and increased N1 response to the self-generated voice. HP did not affect the P2 response to voices. These findings confirm that both prediction and comparison of predicted and perceived feedback to a self-generated voice are altered in individuals with AVH predisposition. Specific alterations in the processing of self-generated vocalizations may establish a core feature of the psychosis continuum.

Action-effect binding and agency.

The sense of agency is a pervasive phenomenon that accompanies conscious acting and extends to the consequences of one's actions in the environment. Subjective feelings of agency are typically explained in terms of predictive processes, based on internal forward models inherent to the sensorimotor system, and postdictive processes, i.e., explicit, retrospective judgments by the agent. Only recently, research has begun to elucidate the link between sense of agency and more basic processes of human action control. The present study was conducted in this spirit and explored the relation between short-term action-effect binding and explicit agency judgments. We found evidence for such a link in that the participants' short-term action-effect binding predicted subsequent agency ratings. This offers a new perspective on the sense of agency, providing an additional mechanism (together with predictive and postdictive processes) that may underlie its formation.

Motor imagery involves predicting the sensory consequences of the imagined movement

Research on motor imagery has identified many similarities between imagined and executed actions at the behavioral, physiological and neural levels, thus supporting their “functional equivalence”. In contrast, little is known about their possible “computational equivalence”—specifically, whether the brain’s internal forward models predict the sensory consequences of imagined movements as they do for overt movements. Here, we address this question by assessing whether imagined self-generated touch produces an attenuation of real tactile sensations. Previous studies have shown that self-touch feels less intense compared with touch of external origin because the forward models predict the tactile feedback based on a copy of the motor command. Our results demonstrate that imagined self-touch is attenuated just as real self-touch is and that the imagery-induced attenuation follows the same spatiotemporal principles as does the attenuation elicited by overt movements. We conclude that motor imagery recruits the forward models to predict the sensory consequences of imagined movements.

Probing transsaccadic correspondence with reverse correlation.

The phenomenon of visual stability is classically explained by an internal forward model predicting the sensory consequences of an eye movement based on efference copy. However, this model cannot explain why some object displacements go undetected, a phenomenon that may depend on a passive prior belief that the world is stable. With reverse correlation, we investigated saccadic suppression of displacement and found that transsaccadic correspondence operates differently depending on the position of the postsaccadic visual target relative to the primary landing position; when the signal falls within the extent of primary saccadic scatter, observers are less able to remap accurately. Furthermore, we observed that the neural representations driving perceptual and saccadic decisions are similar when saccading to a target, but that the representations driving transsaccadic correspondence are different from those driving secondary saccades.

Modulation of sensory prediction error in Purkinje cells during visual feedback manipulations

It is hypothesized that the cerebellum implements a forward internal model that transforms motor commands into predictions about upcoming movements. The predictions are compared with sensory feedback to generate sensory prediction errors critical to controlling movements. The simple spike firing of cerebellar Purkinje cells both lead and lag movement consistent with representations of motor predictions and sensory feedback. This study tests whether this leading and lagging modulation provides the prediction and sensory feedback necessary to compute sensory prediction errors. Two manipulations of the visual feedback are used in rhesus monkeys performing pseudo-random tracking. Consistent with a forward model, delaying the visual feedback demonstrates that the leading simple spike modulation with position error is time-locked to the hand movement. Reducing the feedback shows that the lagged modulation is directly driven by visual inputs. Therefore, Purkinje cell discharge carries both the motor predictions and sensory feedback required of a forward internal model.

Action Prediction Allows Hypothesis Testing via Internal Forward Models at 6 Months of Age.

We propose that action prediction provides a cornerstone in a learning process known as internal forward models. According to this suggestion infants’ predictions (looking to the mouth of someone moving a spoon upward) will moments later be validated or proven false (spoon was in fact directed toward a bowl), information that is directly perceived as the distance between the predicted and actual goal. Using an individual difference approach we demonstrate that action prediction correlates with the tendency to react with surprise when social interactions are not acted out as expected (action evaluation). This association is demonstrated across tasks and in a large sample (n = 118) at 6 months of age. These results provide the first indication that infants might rely on internal forward models to structure their social world. Additional analysis, consistent with prior work and assumptions from embodied cognition, demonstrates that the latency of infants’ action predictions correlate with the infant’s own manual proficiency.

Decomposition of a sensory prediction error signal for visuomotor adaptation.

To accomplish effective motor control, the brain contains an internal forward model that predicts the expected sensory consequence of a motor command. When this prediction is inaccurate, a sensory prediction error is produced which adapts the forward model to make more accurate predictions of future movements. Other types of errors, such as task performance errors or reward, play less of a role in adapting a forward model. This raises the following question: What unique information is conveyed by the sensory prediction error that results in forward model adaptation? sensory prediction errors typically contain both the magnitude and direction of the error, but it is unclear if both components are necessary for adaptation or a single component is sufficient. In this article, we address this by having participants learn to counter a visuomotor rotation, which induces an angular mismatch between movements of the hand and visual feedback. We manipulated the information content of the visual feedback, in the form of a line, which accurately represented only the magnitude (distance), direction, or both magnitude and direction, of the virtual cursor relative to the target. We demonstrate that sensorimotor adaptation does not occur, or is minimal, when feedback is limited to information about the magnitude of an error. In contrast, sensorimotor adaptation is present when feedback is limited only to the direction of an error or when it contains combined direction and magnitude information. This result stands in contrast to current computational models of cerebellar-based sensorimotor adaptation that use error magnitude to drive adaptation. (PsycINFO Database Record

Motor expertise facilitates the accuracy of state extrapolation in perception

Predicting the behavior of objects in the environment is an important requirement to overcome latencies in the sensorimotor system and realize precise actions in rapid situations. Internal forward models that were acquired during motor training might not only be used for efficiently controlling fast motor behavior but also to facilitate extrapolation performance in purely perceptual tasks. In this study, we investigated whether preceding virtual cart-pole balancing training facilitates the ability to extrapolate the virtual pole motion. Specifically, subjects had to report the expected pole orientation after an occlusion of the pole of 900ms duration. We compared a group of 10 subjects, proficient in performing the virtual cart-pole balancing task, to 10 naïve subjects without motor experience in cart-pole balancing task. Our results demonstrate that preceding motor training increases the accuracy of pole movement extrapolation, although extrapolation is not trained explicitly. Additionally, we modelled subjects’ behaviors and show that the difference in extrapolation performance can be explained by individual differences in the accuracy of internal forward models. When subjects are provided with feedback about the true orientation of the pole after the occlusion in a second phase of the experiment, both groups improve rapidly. The results indicate that the perceptual capability to extrapolate the state of the cart-pole system accurately is implicitly trained during motor learning. We discuss these results in the context of shared representations and action-perception transfer.

Structural Learning in a Visuomotor Adaptation Task Is Explicitly Accessible.

Structural learning is a phenomenon characterized by faster learning in a new situation that shares features of previously experienced situations. One prominent example within the sensorimotor domain is that human participants are faster to counter a novel rotation following experience with a set of variable visuomotor rotations. This form of learning is thought to occur implicitly through the updating of an internal forward model, which predicts the sensory consequences of motor commands. However, recent work has shown that much of rotation learning occurs through an explicitly accessible process, such as movement re-aiming. We sought to determine if structural learning in a visuomotor rotation task is purely implicit (e.g., driven by an internal model) or explicitly accessible (i.e., re-aiming). We found that participants exhibited structural learning: following training with a variable set of rotations, they more quickly learned a novel rotation. This benefit was entirely conferred by the explicit re-aiming of movements. Implicit learning offered little to no contribution. Next, we investigated the specificity of this learning benefit by exposing participants to a novel perturbation drawn from a statistical structure either congruent or incongruent with their prior experience. We found that participants who experienced congruent training and test phase structure (i.e., rotations to rotation) learned more quickly than participants exposed to incongruent training and test phase structure (i.e., gains to rotation) and a control group. These results suggest that structural learning in a visuomotor rotation task is specific to previously experienced statistical structure and expressed via explicit re-aiming of movements.

The cerebellum does more than sensory prediction error-based learning in sensorimotor adaptation tasks.

Individuals with damage to the cerebellum perform poorly in sensorimotor adaptation paradigms. This deficit has been attributed to impairment in sensory prediction error-based updating of an internal forward model, a form of implicit learning. These individuals can, however, successfully counter a perturbation when instructed with an explicit aiming strategy. This successful use of an instructed aiming strategy presents a paradox: In adaptation tasks, why do individuals with cerebellar damage not come up with an aiming solution on their own to compensate for their implicit learning deficit? To explore this question, we employed a variant of a visuomotor rotation task in which, before executing a movement on each trial, the participants verbally reported their intended aiming location. Compared with healthy control participants, participants with spinocerebellar ataxia displayed impairments in both implicit learning and aiming. This was observed when the visuomotor rotation was introduced abruptly (experiment 1) or gradually (experiment 2). This dual deficit does not appear to be related to the increased movement variance associated with ataxia: Healthy undergraduates showed little change in implicit learning or aiming when their movement feedback was artificially manipulated to produce similar levels of variability (experiment 3). Taken together the results indicate that a consequence of cerebellar dysfunction is not only impaired sensory prediction error-based learning but also a difficulty in developing and/or maintaining an aiming solution in response to a visuomotor perturbation. We suggest that this dual deficit can be explained by the cerebellum forming part of a network that learns and maintains action-outcome associations across trials.NEW & NOTEWORTHY Individuals with cerebellar pathology are impaired in sensorimotor adaptation. This deficit has been attributed to an impairment in error-based learning, specifically, from a deficit in using sensory prediction errors to update an internal model. Here we show that these individuals also have difficulty in discovering an aiming solution to overcome their adaptation deficit, suggesting a new role for the cerebellum in sensorimotor adaptation tasks.

Sensorimotor predictions and tool use: Hand-held tools attenuate self-touch

Human survival requires quick and accurate movements, both with and without tools. To overcome the sensorimotor delays and noise, the brain uses internal forward models to predict the sensory consequences of an action. Here, we investigated whether these sensory predictions are computed similarly for actions involving hand-held tools and natural hand movements. We hypothesized that the predictive attenuation of touch observed when touching one hand with the other would also be observed for touches applied with a hand-held tool. We first show that when touch is applied to the left index finger with the right index finger, the perceived force sensation is attenuated, only when the fingers are aligned in a manner that simulates direct physical contact and not when a distance of 25cm is introduced between the hands. We then show that touch applied to the left index finger with a tool held in the right hand at a distance of 25cm produces full sensory attenuation, similar to direct finger-to-finger contact. Finally, we show that touch is attenuated only when the tip of the tool is aligned with the receiving left index finger and not when the tip is placed at a distance of 25cm. Collectively, these results suggest that tool use and natural limb movements share the same computational mechanism for sensory predictions. We submit that the brain uses effector-independent forward models: touch is predicted based on the anticipated position of the current effector (i.e., the tip of the tool) rather than the body part per se.

Boosting and Decreasing Action Prediction Abilities Through Excitatory and Inhibitory tDCS of Inferior Frontal Cortex.

Influential theories suggest that humans predict others' upcoming actions by using their own motor system as an internal forward model. However, evidence that the motor system is causally essential for predicting others' actions is meager. Using transcranial direct current stimulation (tDCS), we tested the role of the inferior frontal cortex (IFC), in action prediction (AP). We devised a novel AP task where participants observed the initial phases of right-hand reaching-to-grasp actions and had to predict their outcome (i.e., the goal/object to be grasped). We found that suppression by cathodal (inhibitory) tDCS of the left IFC, but not the left superior temporal sulcus or the right IFC, selectively impaired performance on the AP task, but not on a difficulty-matched control task. Remarkably, anodal (excitatory) tDCS of the left IFC brought about a selective improvement in the AP task. These findings indicate that the left IFC is necessary for predicting the outcomes of observed human right-hand actions. Crucially, our study shows for the first time that down- and up-regulating excitability within the motor system can hinder and enhance AP abilities, respectively. These findings support predictive coding theories of action perception and have implications for enhancement of AP abilities.

Reach-relevant somatosensory signals modulate tactile suppression.

Tactile stimuli on moving limbs are typically attenuated during reach planning and execution. This phenomenon has been related to internal forward models that predict the sensory consequences of a movement. Tactile suppression is considered to occur due to a match between the actual and predicted sensory consequences of a movement, which might free capacities to process novel or task-relevant sensory signals. Here, we examined whether and how tactile suppression depends on the relevance of somatosensory information for reaching. Participants reached with their left or right index finger to the unseen index finger of their other hand (body target) or an unseen pad on a screen (external target). In the body target condition, somatosensory signals from the static hand were available for localizing the reach target. Vibrotactile stimuli were presented on the moving index finger before or during reaching or in a separate no-movement baseline block, and participants indicated whether they detected a stimulus. As expected, detection thresholds before or during reaching were higher compared with baseline. Tactile suppression was also stronger for reaches to body targets than external targets, as reflected by higher detection thresholds and lower precision of detectability. Moreover, detection thresholds were higher when reaching with the left than with the right hand. Our results suggest that tactile suppression is modulated by position signals from the target limb that are required to reach successfully to the own body. Moreover, limb dominance seems to affect tactile suppression, presumably due to disparate uncertainty of feedback signals from the moving limb.NEW & NOTEWORTHY Tactile suppression on a moving limb has been suggested to release computational resources for processing other relevant sensory events. In the current study, we show that tactile sensitivity on the moving limb decreases more when reaching to body targets than external targets. This indicates that tactile perception can be modulated by allocating processing capacities to movement-relevant somatosensory information at the target location. Our results contribute to understanding tactile processing and predictive mechanisms in the brain.

Embodiment and American Sign Language

Little is known about how individual signs that occur in naturally produced signed languages are recognized. Here we examine whether sign understanding may be grounded in sensorimotor properties by evaluating a signer’s ability to make lexical decisions to American Sign Language (ASL) signs that are articulated either congruent with or incongruent with the observer’s own handedness. Our results show little evidence for handedness congruency effects for native signers’ perception of ASL, however handedness congruency effects were seen in non-native late learners of ASL and hearing ASL-English bilinguals. The data are compatible with a theory of sign recognition that makes reference to internally simulated articulatory control signals — a forward model based upon sensory-motor properties of one’s owns body. The data suggest that sign recognition may rely upon an internal body schema when processing is non-optimal as a result of having learned ASL later in life. Native signers however may have developed representations of signs which are less bound to the hand with which it is performed, suggesting a different engagement of an internal forward model for rapid lexical decisions.

Foot placement relies on state estimation during visually guided walking.

Accurate foot placement is essential for safe walking. We used computational models and human walking experiments to test how our nervous system achieves this accuracy. We find that our control of foot placement beneficially combines sensory feedback with internal forward model predictions to accurately estimate the body's state. Our results match recent computational neuroscience findings for reaching movements, suggesting that state estimation is a general mechanism of human motor control.

Pantomime-Grasping: Advance Knowledge of Haptic Feedback Availability Supports an Absolute Visuo-Haptic Calibration.

An emerging issue in movement neurosciences is whether haptic feedback influences the nature of the information supporting a simulated grasping response (i.e., pantomime-grasping). In particular, recent work by our group contrasted pantomime-grasping responses performed with (i.e., PH+ trials) and without (i.e., PH− trials) terminal haptic feedback in separate blocks of trials. Results showed that PH− trials were mediated via relative visual information. In contrast, PH+ trials showed evidence of an absolute visuo-haptic calibration—a finding attributed to an error signal derived from a comparison between expected and actual haptic feedback (i.e., an internal forward model). The present study examined whether advanced knowledge of haptic feedback availability influences the aforementioned calibration process. To that end, PH− and PH+ trials were completed in separate blocks (i.e., the feedback schedule used in our group’s previous study) and a block wherein PH− and PH+ trials were randomly interleaved on a trial-by-trial basis (i.e., random feedback schedule). In other words, the random feedback schedule precluded participants from predicting whether haptic feedback would be available at the movement goal location. We computed just-noticeable-difference (JND) values to determine whether responses adhered to, or violated, the relative psychophysical principles of Weber’s law. Results for the blocked feedback schedule replicated our group’s previous work, whereas in the random feedback schedule PH− and PH+ trials were supported via relative visual information. Accordingly, we propose that a priori knowledge of haptic feedback is necessary to support an absolute visuo-haptic calibration. Moreover, our results demonstrate that the presence and expectancy of haptic feedback is an important consideration in contrasting the behavioral and neural properties of natural and simulated grasping.