Pre-processed motion information has been reported to trigger active exploration responses (tracking, pursuit) or the opposite (avoidance) (1,2). Two thalamic information flows (the Magno and Koniocellular pathways) carry this information, directly to the superior colliculus for rapid reaction and via the cortex where they are integrated. The former thalamic pathway has been broadly studied, whilst the latter lacks understanding. To address this question, we study the responses to motion information in a reduced bio-inspired computational model (10), emulating the thalamo-cortico-collicular visual system of mammals. This model integrates knowledge about the properties of both pathways (6) and is based on known projections between these structures (4,8). Its cortical and collicular maps are implemented with Dynamic Neural Fields (DNF) (7). We study how from such local operators emerge decisions corresponding to complex behaviors (9), such as to approach or to flee. The analysis of the responses of the model to image sequences, resulting from the interplay between the dynamics of the DNFs, yields interesting phenomena. Thanks to the Konio pathway (3), providing raw information about the identity of the stimulus (threat/target), the superior colliculus may choose the upcoming behavior (to flee in presence of a threat), while the cortical mechanisms modulate this rudimentary sensory-motor loop. Such detection is based on two functional interacting ingredients: 1) The capability of the retina to provide coarse, rapid visual event detection (e.g. a looming motion expected to correspond to an approaching target), thus a-priori "assumptions" about the visual surroundings. 2) The algorithmic capability of dual kinds of connectivity of the thalamic neurons (precise/diffuse connectivity standing for core/matrix (5)) to treat, in sequence, different perceptual hypotheses. Our modeling and related numerical experiments demonstrate the functionality of such dual mechanism.