Deep Learning Based Situation Awareness for Multiple Missiles Evasion

TL;DR

This work tackles UAV situational awareness in BVR air combat by extending MD-based risk assessment to multiple incoming missiles using a deep learning framework trained on high-fidelity JSBSim dynamics. It trains single-threat FNN predictors for a predefined set of evasive maneuvers and then aggregates across threats by taking per-maneuver minima to represent the worst-case MD, which is subsequently presented to operators via an intuitive circular visualization. The study demonstrates scenarios with three, four, and six missiles, revealing that some cases yield a safe maneuver (e.g., southward for three missiles) while others offer no safe options (e.g., four missiles at close range). It also shows the method runs in real time and remains informative under sensor uncertainties, providing a practical tool for balancing mission goals against safety in complex, uncertain multi-threat environments.

Abstract

As the effective range of air-to-air missiles increases, it becomes harder for human operators to maintain the situational awareness needed to keep a UAV safe. In this work, we propose a decision support tool to help UAV operators in Beyond Visual Range (BVR) air combat scenarios assess the risks of different options and make decisions based on those. Earlier work focused on the threat posed by a single missile, and in this work, we extend the ideas to several missile threats. The proposed method uses Deep Neural Networks (DNN) to learn from high-fidelity simulations to provide the operator with an outcome estimate for a set of different strategies. Our results demonstrate that the proposed system can manage multiple incoming missiles, evaluate a family of options, and recommend the least risky course of action.

Figures (6)

• Figure 1: Symbolic representation of a situation where the UAV is facing three incoming missiles. The exact current locations of the missiles are unknown, but estimates of the time and location of the launches are available. In Figures \ref{['fig:testnn_path_Esc3']}-\ref{['fig:testnn_path_Esc6']}, the colored fields around the aircraft icon are used to show the predicted miss distance (MD) of an evasive maneuver in that direction. Based on this, the operator can make a trade-off between mission goals and risks when determining what course to choose.
• Figure 2: Fully connected Feed-forward Neural Network with n-dimensional input $\Vec{x}$, neural network weights $w^{k}_{ij}$ and output vector $\Vec{y}$.
• Figure 3: The unmanned F16 faces three incoming missiles, with the starting location indicated by dots. As can be seen, the south-eastern options are safe (green circle segments), while the north-western ones are not (red circle segments). If the mission objectives required a northeastern course, that would have been possible, but with a considerably smaller miss distance (orange circle segments). In this simulation, the operator chose the safest maneuver $\pi_S$, flying south.
• Figure 4: An illustration of the estimated MD distributions, given the sensor variances of Table \ref{['tab:var_obs']}, over time for each flight direction from the scenario in Figure \ref{['fig:testnn_path_Esc3']}. The plotted MD circles in the other figures can be seen as snapshots from a given time instant of this plot. Note that the operator applies $\pi_S$ and thus the estimates for that maneuver remains roughly constant over time. The operator can continuously balance these risk estimates with other mission objectives to decide the proper course of action.
• Figure 5: The unmanned F16$\mathcal{B}$ faces four incoming missiles fired at close range, with dots indicating the starting positions. All policies consistently predict a hit at this range (the circle is red), so there are no safe evasive maneuvers. In such a case, the operator might fire all remaining weapons in a final effort or hope that electronic warfare options such as chaffs or flares might save the UAV in the last second.