The Neural Computation on Animal Collision Detection

Abstract

Promotion: The time-to-collision (TTC) in collision course is a critical neural computation for an animal to response quickly and accurately on collision detection, such as to avoid a looming object. The physical formula of TTC was not feasible for neural computation realization. Our work: We deduce a new approximation of TTC, i.e. 1/τ=Mθ−P/(θ+Q)+N, where θ is the visual angle subtended to the incoming object while M,P,Q and N are constants determined by a practical collision course. Accordingly, we construct a neural computational model (NCM) to carry out the TTC. Results: The simulated results showed that NCM with neurons of sigmoid response can accurately estimate TTC and well reconstruct the practical animal collision experiments. Conclusion: The NCM might be a potential and realizable neural computation mechanism on animal collision detection.