Action observation

Abstract. We report results of an interdisciplinary project which aims at endowing a real robot system with the capacity for learning by goal-directed imitation. The control architecture is biologically inspired as it reflects recent experimental findings in action observation/execution studies. We test its functionality in variations of an imitation paradigm in which the artefact has to reproduce the observed or inferred end state of a grasping-placing sequence displayed by a human model. 1

Research has shown that the brain is constantly making predictions about future events. Theories of prediction in perception, action and learning suggest that the brain serves to reduce the discrepancies between expectation and actual experience, i.e., by reducing the prediction error. Forward models of action and perception propose the generation of a predictive internal representation of the expected sensory outcome, which is matched to the actual sensory feedback. Shared neural representations have been found when experiencing one's own and observing other's actions, rewards, errors, and emotions such as fear and pain. These general principles of the "predictive brain" are well established and have already begun to be applied to social aspects of cognition. The application and relevance of these predictive principles to social cognition are discussed in this article. Evidence is presented to argue that simple non-social cognitive processes can be extended to explain complex cognitive processes required for social interaction, with common neural activity seen for both social and non-social cognitions. A number of studies are included which demonstrate that bottom-up sensory input and top-down expectancies can be modulated by social information. The concept of competing social forward models and a partially distinct category of social prediction errors are introduced. The evolutionary implications of a "social predictive brain" are also mentioned, along with the implications on psychopathology. The review presents a number of testable hypotheses and novel comparisons that aim to stimulate further discussion and integration between currently disparate fields of research, with regard to computational models, behavioral and neurophysiological data. This promotes a relatively new platform for inquiry in social neuroscience with implications in social learning, theory of mind, empathy, the evolution of the social brain, and potential strategies for treating social cognitive deficits.

We have suggested that the mirror-neuron system might be usefully understood as implementing Bayes-optimal perception of actions emitted by oneself or others. To substantiate this claim, we present neuronal simulations that show the same representations can prescribe motor behavior and encode motor intentions during action–observation. These simulations are based on the free-energy formulation of active inference, which is formally related to predictive coding. In this scheme, (generalised) states of the world are represented as trajectories. When these states include motor trajectories they implicitly entail intentions (future motor states). Optimizing the representation of these intentions enables predictive coding in a prospective sense. Crucially, the same generative models used to make predictions can be deployed to predict the actions of self or others by simply changing the bias or precision (i.e. attention) afforded to proprioceptive signals. We illustrate these points using simulations of handwriting to illustrate neuronally plausible generation and recognition of itinerant (wandering) motor trajectories. We then use the same simulations to produce synthetic electrophysiological responses to violations of intentional expectations. Our results affirm that a Bayes-optimal approach provides a principled framework, which accommodates current thinking about the mirror-neuron system. Furthermore, it endorses the general formulation of action as active inference.

The aim in this experiment was to determine whether demonstrations that focus on end-point related information, in this case, ball-trajectory information (BALL), are more effective in teaching motor skills than more traditional demonstrations which focus on movement technique (BODY). Nineteen participants with low-level soccer experience practised a left-footed, soccer-chip shot, which required them to land a ball on a target, while clearing a height barrier. Information concerning how to achieve the task goal was manipulated. Participants either received demonstrations of an expert performing the skill (i.e., BODY, n = 10) or they received a demonstration of the expert's ball flight path (BALL, n = 9). The participants were asked to match the criterion flight or form to achieve the task goal. Feedback concerning ball flight and movement form was controlled, although all participants received KR. Trials were videotaped for analyses and feedback and movement kinematics were colle...

The aim in this experiment was to determine whether demonstrations that focus on end-point related information, in this case, ball-trajectory information (BALL), are more effective in teaching motor skills than more traditional demonstrations which focus on movement technique (BODY). Nineteen participants with low-level soccer experience practiced a left footed, soccer-chip shot, which required them to land a ball on a target, while clearing a height barrier. Information concerning how to achieve the task goal was ...

The uncanny valley hypothesis suggests that robots that are humanoid in appearance elicit positive and empathetic responses, but that there is a point where the robot design is very close to human, the robot becomes repulsive. A possible mechanism underlying this phenomenon is based on the predictive coding theory of neural computations. According to this framework, certain neural systems in the brain can ascribe humanness to a robot that is highly human-like in its appearance, and if the robot’s behavior does not match in realism to the appearance, there will be a processing conflict the neural network will need to resolve. Although this hypothesis is consistent with previous results in the field, empirical work directly testing it is lacking. Here we addressed this gap with a cognitive neuroscience study: We recorded electrical brain activity from the human brain using electroencephalography (EEG) as human subjects viewed images and videos of three agents: A female adult (human), a robot agent closely resembling her (android), and the same robot in a more mechanical appearance (robot). The human and robot had congruent appearance and movement (human with biological appearance and movement; robot with mechanical appearance and movement), and the android had incongruent appearance and movement (biological appearance but mechanical movement). We hypothesized that the android would violate the brain’s predictions since it has a biological appearance, but mechanical movement, whereas the other agents would not lead to such a conflict (robot looks mechanical and moves mechanically; human looks biological and moves biologically). We focused on the N400 ERP component derived from the EEG data. Since the N400 has a greater amplitude for anomalies and violations based on preceding context, we hypothesized the amplitude would be significantly greater for the android in the moving condition than the still condition, whereas the moving and still conditions of the robot and human stimuli would not differ. Our results confirmed out hypothesis, indicating that the uncanny valley might at least partially be due to violations of the brain’s internal predictions about almost-but-not-quite-human robots. Interdisciplinary studies like this one not only allows us to understand the neural basis of human social cognition but also informs robotics about what kind of robots we should design for successful human-robot interaction.

In the present study, we demonstrated that observation of hand rotation had specific facilitation effects on a classical motor imagery task, the hand-laterality judgement. In Experiment 1, we found that action observation improved subjects’ performance on the hand laterality but not on the letter rotation task (stimulus specificity). In Experiment 2, we demonstrated that this facilitation was not due to mere observation of a moving hand, because it was triggered by observation of manual rotation but not of manual prehension movements (motion specificity). In Experiment 3, this stimulus- and motion-specific effect was found to be right hand-specific, compatible with left-hemispheric specialization in motor imagery but not in action observation. These data provided direct support to the idea that different simulation states, such as action observation and motor imagery, share some common mechanisms but also show specific functional differences.