AN INSECT-INSPIRED TARGETING/EVASION REFLEX FOR AUTONOMOUS AIR VEHICLES Ravi Vaidyanathan 1,3* * , Charles A. Williams 1 , Troy S. Prince 1 Roy E.Ritzmann 2 , and Roger D. Quinn 3 1 Orbital Research Inc., Cleveland, OH, USA 2 Department of Biology, Case Western Reserve University, Cleveland OH, USA 3 Department of Mechanical Engineering, Case Western Reserve University, Cleveland OH, USA * Corresponding author: raviv@orbitalresearch.com Abstract This paper investigates a biologically inspired target seeking reflex for the endgame phase of autonomous munition flight. The reflex is based upon an artificial neural network model of the American Cockroach’s escape reflex, and combines exteroceptive and proprioceptive inputs to produce output commands to a Linear Quadratic Regulator (LQR) autopilot that guides the munition to an optimal path destination and orientation for target strike. Simulation and flight test results are presented that demonstrate the reflex’s capability for instantaneous target strike on evasive targets, even in the presence of false or disruptive sensor data. I. Background The problem of directing a tactical missile to intercept mobile targets has been referred to as the most challenging of guidance and control problems [1]. In the classical approach, known as proportional navigation (Pronav), a controller attempts to align the velocity vector of the munition with a line-of-sight vector to its target. Even today, Pronav provides the basis for much of munition guidance and control [2]. In the air-to-air guidance scenario, three fundamental phases of munition flight have been defined [3]. These phases are commonly referred to as midcourse, terminal, and endgame stages of flight. Midcourse guidance is, in effect, from the time of launch until target sensor acquisition. Once the sensors acquire the target, terminal guidance is initiated. The last second of terminal guidance is referred to as endgame. Endgame is worth treating as a separate problem since uncertainties in guidance need to be corrected much more rapidly, thrust may be unreliable due to time delay [2], and missile failure is most often associated with this phase [1]. The endgame part of intercept has received less attention in guidance and control literature than its midcourse and terminal counterparts. Cottrell [4] attempted to improve end-game performance by extending classical Pronav, Dowdle et al., [5], generalized the LQG regulator, Looze et al. [6] used roll commands to compensate for target estimation error. Cho et al. [7] proposed drag minimization for missiles with non-constant velocities. Forte and Shinar [8] formulated the planar intercept as a differential game. Dougherty and Speyer [9] concluded that integrating air frame response equations is typically not feasible in real-time, and proposed pulse functions to approximate forces. It has also been noted that although non-linear models could aid in air vehicle control they are typically too large for on-board computers [1,3]. Several researchers have proposed neural networks for aircraft usage due to their capability to represent complex data in compact structures [10]. II. Introduction In this work, a targeting/goal-acquisition reflex for autonomous air vehicles is introduced based upon a distributed network of artificial neurons that mimic the neural organization of the escape system in the American cockroach. Although the escape response of the American Cockroach has evolved under a set of goals that are obviously different from that of a target- seeking reflex [11], extracting certain aspects of its performance nevertheless has the potential to improve endgame munition guidance. The primary deviation in functionality between an intercept and escape response lies in the fact that the intercept problem is