One-to-one Air-combat Maneuver Strategy Based on Improved TD3 Algorithm

In this paper, the problem of one-to-one unmanned fighter air-combat autonomous maneuver decision including missile evasion is considered. Firstly, the model of one-to-one unmanned fighter air-combat is established by reducing the 3-dimensional space into the 2-dimensional planes. Considering that Twin Delayed Deep Deterministic policy gradient algorithm (TD3) can deal with continuous control problems and has strong robustness, an improved TD3 algorithm is presented to realize the air-combat autonomous maneuver decision. To increase training efficiency, a new loss term between recommended action and actor output is proposed in this algorithm. Finally, one-to-one air-combat is simulated on DCS World platform and the effectiveness of this algorithm is verified.