Researchers from a team led by Zhen Yang and colleagues have proposed an intelligent terminal guidance law designed to improve the ability of air-to-air missiles to intercept highly maneuverable targets. The study, published in the journal Neural Computing and Applications, proposes a method based on the Twin Delayed Deep Deterministic Policy Gradient, or TD3, algorithm. This approach aims to address specific limitations found in traditional terminal guidance laws, which often struggle with poor estimation of target maneuvers and difficulty handling complex combat environments.

The core of the proposed system models the guidance process as a Markov Decision Process, or MDP.

To enhance the missile’s ability to anticipate the target’s movements, the team developed a trajectory prediction algorithm based on a bidirectional long-short-term memory, or Bi-LSTM, network. This neural network component is responsible for estimating the future states of the target. These predicted states are then incorporated into the state space of the reinforcement learning agent.

The study also introduces a heuristic reward shaping strategy to improve the training process of the reinforcement learning agent. The researchers designed this specific strategy to accelerate policy learning and improve the stability of convergence.

Furthermore, the team implemented a progressive curriculum learning training strategy to further enhance training efficiency. This method involves gradually increasing the complexity of the tasks during the training phase.

The simulation results presented in the study demonstrate that the proposed intelligent guidance law enables the missile to effectively evade incoming threats and accurately intercept highly maneuverable targets. The authors report that the system shows excellent adaptability and robustness in complex engagement scenarios.

While the study provides promising simulation results, it is important to note that the findings are based on simulated engagement scenarios.

Subject of Research: Development of a deep reinforcement learning-based terminal guidance law for air-to-air missiles using TD3 and Bi-LSTM networks.

Article Title: Deep reinforcement learning based intelligent terminal guidance for intercepting highly maneuverable target

Article References: Yang, Z., Li, S., Liu, Y., Wang, H., Yuwen, S., & Zhou, D. (2026). Deep reinforcement learning based intelligent terminal guidance for intercepting highly maneuverable target. Neural Computing and Applications, 38(21), Article 880. https://doi.org/10.1007/s00521-026-12584-0

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12584-0

Keywords: deep reinforcement learning, terminal guidance, air-to-air missile, TD3 algorithm, Bi-LSTM network, trajectory prediction, Markov Decision Process, aerospace control, intelligent systems, Neural Computing and Applications, Deep, reinforcement