Neural Reflex Networks for Automating Quadcopter Drone Obstacle Avoidance

Abstract

The physically modelled neural and nervous networks pioneered more than two decades ago by Mark Tilden, E. A. Rietman and collaborators, M. Ashkenazi et al. have proven to be a robust and interesting way to obtain powerful emergent behavior by utilizing neuromimetic circuitry. Using a physical representation of biologic neurons, both motor (NU) and cortical (NV) these structures mimic simple reflex arcs present in a large number of evolved organisms. The simple circuits using logic gate oscillators wired as integrators or pulse delay loops with sensors coupled as current injectors or variable resistors of different types demonstrated unexpected emergent „survival” behaviors when connected in chains or loops of several neurons. Mark Tilden calls the simplest functional unit of such looped structures „bicores”-as two neurons linked together in a loop already generate meaningful behavior when their inputs are linked to appropriate sensors. These powerful neuromimetic machines allow for a robust implementation of automated responses in autonomous or semi-autonomous robots. Quadcopters are a very good target for neural network control/stabilization because of their unique flight dynamics and normal control procedures. Obstacle avoidance and stabilization are simple tasks for a well-tuned physical neural network.