Learning a complex motor skill from video and point-light demonstrations

The aim of this study was to compare the learning process of a highly complex ballet skill following demonstrations of point-light and video models. 16 participants divided into point-light and video groups (ns = 8) performed 160 trials of a pirouette, equally distributed in blocks of 20 trials, alternating periods of demonstration and practice, with a retention test a day later. Measures of head and trunk oscillation, coordination disparity from the model, and movement time difference showed similarities between video and point-light groups; ballet experts' evaluations indicated superiority of performance in the video over the point-light group. Results are discussed in terms of the task requirements of dissociation between head and trunk rotations, focusing on the hypothesis of sufficiency and higher relevance of information contained in biological motion models applied to learning of complex motor skills.     

Perception of American sign language in dynamic point-light displays

Perception of dynamic events of American Sign Language (ASL) was studied by isolating information about motion in the language from information about form. Four experiments utilized Johansson's technique for presenting biological motion as moving points of light. In the first, deaf signers were highly accurate in matching movements of lexical signs presented in point-light displays to those normally presented. Both discrimination accuracy and the pattern of errors were similar in this matching task to that obtained in a control condition in which the same signs were always represented normally. The second experiment showed that these results held for discrimination of morphological operations presented in point-light displays as well. In the third experiment, signers were able to accurately identify signs of a constant handshape and morphological operations acting on signs presented in point-light displays. Finally, in Experiment 4, we evaluated what aspects of the motion patterns carried most of the information for sign identifiability. We presented signs in point-light displays with certain lights removed and found that the movement of the fingertips, but not of any other pair of points, is necessary for sign identification and that, in general, the more distal the joint, the more information its movement carries.     

Perception of biological motion: a stimulus set of human point-light actions.

We present a set of stimuli representing human actions under point-light conditions, as seen from different viewpoints. The set contains 22 fairly short, well-delineated, and visually "loopable" actions. For each action, we provide movie files from five different viewpoints as well as a text file with the three spatial coordinates of the point lights, allowing researchers to construct customized versions. The full set of stimuli may be downloaded from www.psychonomic.org/archive/.     

Visual search and coordination changes in response to video and point-light demonstrations without KR

The authors examined the observational learning of 24 participants whom they constrained to use the model by removing intrinsic visual knowledge of results (KR). Matched participants assigned to video (VID), point-light (PL), and no-model (CON) groups performed a soccer-chipping task in which vision was occluded at ball contact. Pre- and posttests were interspersed with alternating periods of demonstration and acquisition. The authors assessed delayed retention 2-3 days later. In support of the visual perception perspective, the participants who observed the models showed immediate and enduring changes to more closely imitate the model's relative motion. While observing the demonstration, the PL group participants were more selective in their visual search than were the VID group participants but did not perform more accurately or learn more.     

Perception of biological motion: A stimulus set of human point-light actions

We present a set of stimuli representing human actions under point-light conditions, as seen from different viewpoints. The set contains 22 fairly short, well-delineated, and visually “loopable” actions. For each action, we provide movie files from five different viewpoints as well as a text file with the three spatial coordinates of the point lights, allowing researchers to construct customized versions. The full set of stimuli may be downloaded fromwww.psychonomic.org/archive/.     

Gender recognition from point-light walkers

Point-light displays of human gait provide information sufficient to recognize the gender of a walker and are taken as evidence of the exquisite tuning of the visual system to biological motion. The authors revisit this topic with the goals of quantifying human efficiency at gender recognition. To achieve this, the authors first derive an ideal observer for gender recognition on the basis of center of moment (J. E. Cutting, D. R. Proffitt, & L. T. Kozlowski, 1978) and, with the use of anthropometric data from various populations, show optimal recognition of approximately 79% correct. Next, they perform a meta-analysis of 21 experiments examining gender recognition, obtaining accuracies of 66% correct for a side view and 71% for other views. Finally, results of the meta-analysis and the ideal observer are combined to obtain estimates of human efficiency at gender recognition of 26% for the side view and 47% for other views.     

Recognizing the sex of a walker from a dynamic point-light display

The sex of human walkers can be recognized without familiarity cues from displays of pointlight sources mounted on major joints. Static versions of these abstract displays do not permit accurate recognition of sex. Variation in the degree of armswing or in walking speed generally interferes with recognition, except that faster speeds are associated somewhat with improved recognition of females. Lights on upper-body joints permit more accurate guesses than do Lights on lower-body joints, but identification is possible even from minimal displays, with lights placed only on the ankles. No feedback was given to observers. Confidence judgments of sex relate to the accuracy of responses in a manner that suggests that viewers know what they are doing.     

Audience-contingent variation in action demonstrations for humans and computers

People may exhibit two kinds of modifications when demonstrating action for others: modifications to facilitate bottom-up, or sensory-based processing; and modifications to facilitate top-down, or knowledge-based processing. The current study examined actors' production of such modifications in action demonstrations for audiences that differed in their capacity for intentional reasoning. Actors' demonstrations of complex actions for a non-anthropomorphic computer system and for people (adult and toddler) were compared. Evidence was found for greater highlighting of top-down modifications in the demonstrations for the human audiences versus the computer audience. Conversely, participants highlighted simple perceptual modifications for the computer audience, producing more punctuated and wider ranging motions. This study suggests that people consider differences in their audiences when demonstrating action.     

Movement kinematics affect action prediction: comparing human to non-human point-light actions

The influence of movement kinematics on the accuracy of predicting the time course of another individual's actions was studied. A human point-light shape was animated with human movement (natural condition) and with artificial movement that was more uniform regarding velocity profiles and trajectories (artificial condition). During brief occlusions, the participants predicted the actions in order to judge after occlusion whether the actions were continued coherently in time or shifted to an earlier or later frame. Error rates and reaction times were increased in the artificial compared to the natural condition. The findings suggest a perceptual advantage for movement with a human velocity profile, corresponding to the notion of a close interaction between observed and executed movement. The results are discussed in the framework of the simulation account and alternative interpretations are provided on the basis of correlations between the velocity profiles of natural and artificial movements with prediction performance.     

The informational properties of the throwing arm for anticipation of goal-directed action

We examined the informational value of biological motion from the arm in predicting the location of a thrown ball. In three experiments, participants were classified as being skilled and less skilled based on their actual performance on the task (i.e., using a within-task criterion). We then presented participants with a range of stick figure representations and required them to predict throw direction. In Experiment 1, we presented stick figure movies of a full body throwing action, right throwing arm plus left shoulder and throwing arm only. Participants were able to anticipate throw direction above chance under all conditions irrespective of perceptual skill level, with the perceptually skilled participants excelling under full body conditions. In Experiment 2, we neutralized dynamical differences in motion to opposing throw directions from the shoulder, elbow and wrist of the throwing arm. Neutralizing the wrist location negatively affected anticipation performance in all participants reducing accuracy to below chance. In Experiment 3, we presented movies of the motion wrist location alone and the upper section of the throwing arm (shoulder-elbow). Participants were able to successfully anticipate above chance in these latter two conditions. Our findings suggest that motion of the throwing arm contains multiple sources of information that can help facilitate the anticipation of goal-directed action. Perceptually skilled participants were superior in extracting informational value from motion at both the local and global levels when compared to less skilled counterparts.     

Perceptual coupling of multiple point-light figures

When confronted with multiple bistable stimuli at the same time, the visual system tends to generate a common interpretation for all stimuli. We exploit this perceptual-coupling phenomenon to investigate the perception of depth in bistable point-light figures. Observers indicate the global depth orientation of simultaneously presented point-light figures while the similarity between the stimuli is manipulated. In a first experiment, a higher occurrence of coupled percepts is found for identical figures, but coupling breaks down when either the movement pattern or both the viewpoint and the phase-relation are changed. A second experiment confirms these results, but also demonstrates that different point-light actions can be subject to perceptual coupling as long as they share the same viewpoint and exhibit equivalent degrees of perceptual ambiguity. The data are consistent with an explanation in terms of differential contributions of stored, view-dependent object and action representations and of an interaction between stages processing local stimulus features. The consequences of these results are discussed in the framework of an explicit model for the perception of depth in biological motion.     

Perception in action: multiple roles of sensory information in action control

Perceptual processes play a central role in the planning and control of human voluntary action. Indeed, planning an action is a sensorimotor process operating on sensorimotor units, a process that is based on anticipations of perceptual action effects. I discuss how the underlying sensorimotor units emerge, and how they can be employed to tailor action plans to the goals at hand. I also discuss how even a single action can induce sensorimotor binding, how intentionally implemented short-term associations between stimuli and responses become autonomous, how feature overlap between stimulus events and actions makes them compatible, and why action plans are necessarily incomplete.

Hierarchical Deep Q-Network from imperfect demonstrations in Minecraft

We present Hierarchical Deep Q-Network (HDQfD) that won first place in the MineRL competition. The HDQfD works on imperfect demonstrations and utilizes the hierarchical structure of expert trajectories. We introduce the procedure of extracting an effective sequence of meta-actions and subgoals from the demonstration data. We present a structured task-dependent replay buffer and an adaptive prioritizing technique that allow the HDQfD agent to gradually erase poor-quality expert data from the buffer. In this paper, we present the details of the HDQfD algorithm and give the experimental results in the Minecraft domain.     

Inferring intentions from biological motion: A stimulus set of point-light communicative interactions

We present the first database of communicative interactions reproduced through point-light displays (Communicative Interaction Database). The database contains 20 communicative interactions performed by male and by female couples. For each action, we provide movie files from four different viewpoints, as well as text files with the 3-D spatial coordinates of the point lights, allowing researchers to construct customized versions. By including various types of actions performed with different social motives, the database contains a diverse sample of nonconventional communicative gestures. Normative data collected to assess the recognizability of the stimuli suggest that, for most action stimuli, information in point-light displays is sufficient for clear recognition of the action as communicative, as well as for identification of the specific communicative gesture performed by the actor. The full set of stimuli may be downloaded from http://brm.psychonomic-journals.org/ content/supplemental and from http://ppw.kuleuven.be/labexppsy/lepSite/resources/CID.rar.     

An anxiety-induced bias in the perception of a bistable point-light walker

Human sensitivity for social cues is exquisite, as illustrated by the ease with which simplified point-light movements invoke social and emotional responses. Compared to faces, these biological motion stimuli only recently started to be used to explore questions regarding social cognition and anxiety. We presented human point-light walkers that could be perceived as facing towards or facing away from the observer, and tested whether participants with high social anxiety would perceive these bistable stimuli differently, because this type of stimuli has particular relevance for them. The results showed that observers with high social anxiety tended to see walkers as facing away more frequently than those with low social anxiety. This may mean that high socially anxious observers are biased towards the more positive perceptual alternative because they are motivated to protect themselves against threatening social experiences, but we also explore alternative explanations. The findings are in line with the evidence for a positivity bias in perception, also called wishful seeing, but in contrast with the attentional negativity bias often found in social anxiety. We discuss reasons for this divergence and possible limitations of the current study.     

The perception of emotion from body movement in point-light displays of interpersonal dialogue

We examined whether it is possible to identify the emotional content of behaviour from point-light displays where pairs of actors are engaged in interpersonal communication. These actors displayed a series of emotions, which included sadness, anger, joy, disgust, fear, and romantic love. In experiment 1, subjects viewed brief clips of these point-light displays presented the right way up and upside down. In experiment 2, the importance of the interaction between the two figures in the recognition of emotion was examined. Subjects were shown upright versions of (i) the original pairs (dyads), (ii) a single actor (monad), and (iii) a dyad comprising a single actor and his/her mirror image (reflected dyad). In each experiment, the subjects rated the emotional content of the displays by moving a slider along a horizontal scale. All of the emotions received a rating for every clip. In experiment 1, when the displays were upright, the correct emotions were identified in each case except disgust; but, when the displays were inverted, performance was significantly diminished for some emotions. In experiment 2, the recognition of love and joy was impaired by the absence of the acting partner, and the recognition of sadness, joy, and fear was impaired in the non-veridical (mirror image) displays. These findings both support and extend previous research by showing that biological motion is sufficient for the perception of emotion, although inversion affects performance. Moreover, emotion perception from biological motion can be affected by the veridical or non-veridical social context within the displays.     

On the depth reversibility of point-light actions

We investigate the occurrence of perspective reversals for a depth-ambiguous point-light figure. In addition, we exploit the phenomenon of reversibility to search for stimulus features relevant in the process of depth assignment. Experiment 1 shows that perceptual switches indeed occur during prolonged viewing, although the switches occur infrequently. The reversibility is confirmed in Experiment 2, in which the perceptual ambiguity of the point-light action is manipulated as well as observers’ intention to perceive a particular alternative. In addition, the pattern of eye movements reveals local stimulus features specifically associated with the perception of the different alternatives. In Experiment 3, the importance of these features as determining factors of the initial interpretation is investigated by manipulating the location of the first fixation on the stimulus. Implications for a better understanding of biological motion perception are discussed.     

Emotion perception from dynamic and static body expressions in point-light and full-light displays

Research on emotion recognition has been dominated by studies of photographs of facial expressions. A full understanding of emotion perception and its neural substrate will require investigations that employ dynamic displays and means of expression other than the face. Our aims were: (i) to develop a set of dynamic and static whole-body expressions of basic emotions for systematic investigations of clinical populations, and for use in functional-imaging studies; (ii) to assess forced-choice emotion-classification performance with these stimuli relative to the results of previous studies; and (iii) to test the hypotheses that more exaggerated whole-body movements would produce (a) more accurate emotion classification and (b) higher ratings of emotional intensity. Ten actors portrayed 5 emotions (anger, disgust, fear, happiness, and sadness) at 3 levels of exaggeration, with their faces covered. Two identical sets of 150 emotion portrayals (full-light and point-light) were created from the same digital footage, along with corresponding static images of the 'peak' of each emotion portrayal. Recognition tasks confirmed previous findings that basic emotions are readily identifiable from body movements, even when static form information is minimised by use of point-light displays, and that full-light and even point-light displays can convey identifiable emotions, though rather less efficiently than dynamic displays. Recognition success differed for individual emotions, corroborating earlier results about the importance of distinguishing differences in movement characteristics for different emotional expressions. The patterns of misclassifications were in keeping with earlier findings on emotional clustering. Exaggeration of body movement (a) enhanced recognition accuracy, especially for the dynamic point-light displays, but notably not for sadness, and (b) produced higher emotional-intensity ratings, regardless of lighting condition, for movies but to a lesser extent for stills, indicating that intensity judgments of body gestures rely more on movement (or form-from-movement) than static form information.     

Techniques for the production of point-light and fully illuminated video displays from identical recordings

Illumination of only a few key points on a moving human body or face is enough to convey a compelling perception of human motion. A full understanding of the perception ofbiological motion from point-light displays requires accurate comparison with the perception of motion in normal, fully illuminated versions of the same images. Traditionally, these two types of stimuli (point-light and fully illuminated) have been filmed separately, allowing the introduction of uncontrolled variation across recordings. This is undesirable for accurate comparison of perceptual performance across the two types of display. This article describes simple techniques, using proprietary software, that allow production of point-light and fully illuminated video displays from identical recordings. These techniques are potentially useful for many studies of motion perception, by permitting precise comparison of perceptual performances across point-light displays and their fully illuminated counterparts with accuracy and comparative ease.     

Perception of objects oriented downward from a vertical position

In the present work we investigated people’s perceptions of orientation for surfaces that are conceived of as being sloped downward from vertical against a vertical reference frame. In the three conditions of Experiment 1, participants either (1) placed a ladder against a wall at what they thought was the most stable position, and then estimated its orientation; (2) gave a verbal (conceptual) estimate of what the most stable position of a ladder leaned against a wall would be; or (3) drew a line representing the most stable position of a ladder to be placed against a wall, and then gave a verbal estimate of the ladder’s orientation. Ladder placement was shallower than the most stable position, as were the verbal estimations of both the positioned and drawn orientations and the verbal (conceptual) estimates of the most stable position for a ladder to be leaned against a wall, relative to the actual orientations. In Experiment 2, participants verbally estimated various ladder orientations. The estimates were again shallower than the actual orientations. For orientations between 60° and 90°, the estimates showed a scale compression effect from horizontal. This perceived exaggeration of the orientation of an object typically oriented down from vertical is similar to the perceived exaggeration of the orientation of hills and ramps, typically thought of as oriented up from horizontal. This may point to a generic perceived exaggeration of slant whose direction depends on the conceptual or actual reference frame being used.