Physics-constrained mixture density networks for the inverse problem of Risley prism beam steering

Area of Science:

• Optics and Photonics
• Computational Physics
• Machine Learning Applications

Background:

• Risley prism scanners face challenges with non-unique and unstable inverse mapping, especially under non-paraxial conditions.
• Accurate inverse mapping is crucial for precise beam steering in optical systems.

Purpose of the Study:

• To develop a robust and fast inverse mapping method for Risley prism beam steering.
• To enable direct prediction of prism angles from screen coordinates without paraxial assumptions.

Main Methods:

• A physics-constrained mixture density network was employed for the inverse problem.
• The model utilizes a multi-modal von Mises mixture and embeds a high-precision forward ray model.
• A one-step Gauss-Newton/Levenberg-Marquardt refinement was applied.

Main Results:

• >94% correct-solution identification in simulations.
• Micrometer-level reprojection Root Mean Square Error (RMSE).
• Millisecond-level inference times with enhanced robustness to boundary conditions and moderate noise.

Conclusions:

• The proposed method offers a feasible solution for real-time, high-precision pointing control in Risley prism systems.
• Demonstrates improved accuracy and robustness compared to traditional methods.