Design of Fuzzy Neural Network Guidance Law Based on TakagiSugeno for UAV

Yibo Li, Zitong Li, Yi Wang, Wei ChenInternational Journal of Control and Automation 2014 · DOI

Abstract

Guidance of unmanned aerial vehicles (UAVs) in three-dimensional space is a critical challenge in modern aerospace engineering. This paper presents a three-dimensional fuzzy neural network guidance law based on the Takagi-Sugeno (T-S) fuzzy rules and the proportional guidance principle. By decomposing the 3D tracking trajectory into two separate two-dimensional planes, the system accounts for target mobility and environmental constraints. The fuzzy neural network takes the approaching velocity, vertical line-of-sight (LOS) angular velocity, and horizontal LOS angular velocity as inputs, outputting yaw and pitch acceleration commands. Simulation results against targets in uniform straight-line, curved, and circular motion demonstrate that the proposed law is superior to traditional methods, providing smoother trajectories and better response to maneuvering targets.

Introduction

Guidance laws are the mathematical logic that dictates how a vehicle steers to intercept or track a target. While Proportional Navigation Guidance (PNG) has been the industry standard for decades due to its simplicity, it often struggles with highly maneuverable targets or produces oscillatory trajectories that can stress a UAV's airframe.

Unlike missiles, which prioritize minimal miss distance for a direct hit, UAVs often need to maintain a stable tracking position or reach a specific "attack zone" to deploy sensors effectively. This requires a guidance law that is not just accurate, but also smooth and adaptable to unpredictable movements in three-dimensional space.

Methodology

The researchers combined fuzzy logic with neural networks to create a hybrid controller utilizing the Takagi-Sugeno (T-S) fuzzy model. Unlike traditional Mamdani fuzzy systems that use descriptive linguistic outputs, T-S models use mathematical functions, making them easier to optimize using neural network learning algorithms.

Key aspects of the methodology include:

  • Dimensional Decomposition: The 3D tracking problem is split into horizontal (yaw) and vertical (pitch) planes to simplify the control logic.
  • 5-Layer Fuzzy Neural Network: The architecture consists of an input layer, a fuzzification layer (using Gaussian membership functions), a rule layer with 49 fuzzy rules, a normalization layer, and an output layer.
The 5-layer ANFIS model structure used to implement the Takagi-Sugeno fuzzy inference system.
  • Input/Output Mapping: The network processes range rate (approaching velocity) and LOS angular velocities to generate acceleration commands. Separate fuzzy controllers were established for each plane to account for the asymmetric effects of gravity and aerodynamics.

Key Results and Insights

The authors compared their T-S fuzzy neural network against traditional 3D proportional navigation guidance ($N=5$) across several scenarios, including stationary and highly maneuvering targets.

One of the most significant findings is that the T-S fuzzy neural network naturally produces smoother trajectories than PNG. In aviation, "smoothness" translates to less wear and tear on actuators and more stable data from onboard sensors. The 49 fuzzy rules allow the system to handle edge cases—such as when the UAV is misaligned on one axis—more gracefully than a purely linear controller.

Simulation of the UAV 3D tracking trajectory showing a direct, non-oscillatory path toward the target.

Visual analysis of the simulation plots confirms that the fuzzy neural network produces a much more direct and less "jittery" path toward the target compared to the baseline. The system successfully closed the distance to the target within the 600-meter success threshold across all four simulation scenarios, proving its robustness against maneuvering threats.

Conclusion

This research demonstrates that a hybrid fuzzy neural network approach can significantly improve the tracking performance and structural longevity of UAVs. While the current results are based on kinematic simulations, they provide a strong foundation for future hardware-in-the-loop (HIL) testing. Future work will focus on integrating real-world factors such as sensor noise and atmospheric turbulence to further validate the law's effectiveness in field conditions.

Summary updated

Abstract

Guidance of unmanned aerial vehicle (UAV) in three-dimensional space was studied in this paper. We designed a three-dimensional fuzzy neural network guidance law with the fuzzy rules of Takagi-Sugeno that based on proportional guidance principle and put the characteristics of the fuzzy neural network control into use. We decomposed the 3-D tracking trajectory into two two-dimensional planes and guided respectively, and took the influence of target mobility for guidance trajectory into consider. The design of guidance law was using the principle of proportional guidance, based on the Takagi-Sugeno type fuzzy inference, we took horizontal line of approaching velocity of missile-to-target, vertical line of sight angle velocity and sight angle velocity as inputs of fuzzy neural network, and yaw and pitch acceleration commands of UAV are the outputs, then we simulated the seeker tracking process against the target doing uniform motion in a straight line, curve movement and uniform circular motion respectively for UAV. Verified by simulation in this paper, the effect of the design of guidance law is ideal, and the proposed law is superior to the others and has good effect in response to the maneuvering target, and guidance trajectory is smoother.
BibTeX
@article{li2014design,
  author = {Yibo Li and Zitong Li and Yi Wang and Wei Chen},
  title = {{Design of Fuzzy Neural Network Guidance Law Based on TakagiSugeno for UAV}},
  journal = {International Journal of Control and Automation},
  year = {2014},
  doi = {10.14257/ijca.2014.7.12.13},
}
Figure