Bioinspired approach-sensitive neural network for collision detection in cluttered and dynamic backgrounds

Rapid, accurate and robust detection of looming objects in cluttered moving backgrounds is a significant and challenging problem encountered when designing robotic visual systems to perform collision detection and avoidance tasks. Inspired by the neural circuit involved in the elementary motion vision of the mammalian retina, this study proposes a bioinspired approach-sensitive neural network (ASNN). The three main contributions of this work are as follows. First, a direction-selective visual processing module is built based on the spatiotemporal energy framework, which can estimate motion direction accurately via only two mutually perpendicular spatiotemporal filtering channels. Second, a novel approach-sensitive neural network is modeled as a push–pull structure formed by ON and OFF pathways, which responds strongly to approaching motion, but is insensitive to lateral motion. Finally, a method of direction-selective inhibition is introduced, which is able to effectively suppress translational backgrounds. Extensive synthetic and real robotic experiments indicate that the proposed model is able to not only detect collisions accurately and robustly in cluttered and dynamic backgrounds but also extract additional information on the approaching object, such as position and direction, which is critical for guiding rapid decision making.