Direction Selectivity in Naturalistic Action Observation: Distributed Representations Across the Action Observation Network

Perceiving the direction of observed actions is critical for interpreting intentions and guiding social interaction. While direction selectivity has been extensively studied with simple stimuli such as dots, gratings, or point-light displays (PLDs), little is known about how the brain encodes direction in naturalistic, repetitive actions that are seen frequently in daily life. The present fMRI study investigated direction-selective representations during observation of complex actions performed along three bidirectional dimensions (left-right, up-down, front-back) within a 96-video stimulus set. The brain activity was analyzed using multivariate pattern analysis (MVPA) and multiple regression representational similarity analysis (RSA). MVPA revealed above-chance classification of action direction across occipital, parietal, and motor cortices, with the highest decoding in occipital, primary motor, and somatosensory regions. Crucially, RSA demonstrated that when accounting for low-level and motor features, direction information was still represented in early visual cortex, occipito-temporal areas, parietal regions, and motor-related regions. These findings indicate that action direction is represented across multiple levels of the action observation network (AON), extending from early sensory regions to higher-order parietal and frontal cortices. By using naturalistic, repetitive action videos, this study provides new evidence that the coding of action direction in the human brain is broadly distributed, reflecting the complexity of perceiving actions in everyday life. These findings suggest that direction selectivity is a core feature of the action observation network, linking basic motion processing with higher-level action understanding.