TURBO-RL: turbulence mitigation using reinforcement learning for severe optical aberrations.

Area of Science:

• Optical engineering
• Computational imaging
• Machine learning applications

Background:

• Optical distortions and aberrations limit high-resolution imaging in critical fields like astronomy and remote sensing.
• Existing wavefront sensing methods struggle with severe atmospheric turbulence and low photon counts.
• Adaptive optics systems are effective for mild turbulence but fail in extreme conditions.

Purpose of the Study:

• To develop a novel method for estimating and correcting wavefront errors caused by severe optical turbulence.
• To enable high-resolution imaging in challenging environments with limited light.
• To overcome the limitations of traditional wavefront sensors in extreme turbulence.

Main Methods:

• Implementation of TURBO-RL (TURBulence mitigatiOn using Reinforcement Learning), a system utilizing reinforcement learning and a convolutional neural network.
• Employing a single deformable mirror for wavefront error estimation and correction.
• Testing performance in severe turbulence conditions (D/r0=100) with low photon counts (approx. 590 photons).

Main Results:

• Successful estimation and correction of wavefront errors in severe turbulence.
• Demonstrated capability for guide star imaging under extreme conditions (D/r0=100) with minimal photons.
• TURBO-RL shows potential to outperform existing methods in low-light, high-turbulence scenarios.

Conclusions:

• TURBO-RL offers a promising solution for high-resolution imaging in severe optical turbulence.
• The method's efficiency with low photon counts makes it suitable for challenging astronomical and remote sensing applications.
• This approach may provide a more compact and cost-effective alternative to traditional wavefront sensors.