On-sky demonstration of reinforcement learning for adaptive optics control
Jalo Nousiainen, Vincent Chambouleyron, Benoit Neichel, Sylvain Cetre, Jean-Francois Sauvage, Angelie Alagao, Markus Kasper, Jonathan Dray, Romain Fetick, Byron Engler
THE PROBLEM
This paper focuses on Imitation & Reinforcement LearningReinforcement Learning (RL)Teaching a robot through trial and error using rewards.. This is the first real-world Simulation & Sim-to-RealDeploymentPutting the trained system on a real robot. of RL-based adaptive optics Control & PlanningControlThe method used to make the robot move the way you want. on an actual telescope, proving that learned Control & PlanningControlThe method used to make the robot move the way you want. policies can outperform traditional integrator controllers while automatically handling vibrations and Data, Distributions & Training IssuesNoiseUnwanted variation or randomness in sensor readings or actuation.. For roboticists, it validates that Imitation & Reinforcement LearningReinforcement Learning (RL)Teaching a robot through trial and error using rewards. can scale from Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested. to high-frequency real-world Control & PlanningControlThe method used to make the robot move the way you want. tasks with tight Simulation & Sim-to-RealLatencyDelay between input, computation, and action. budgets and changing conditions—without per-condition retuning. Read the paper by tracking the Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. definition, the Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. or data assumptions, and the evidence that supports the claimed improvement.
HOW IT WORKS
Task framing
Core method
Data and supervision
Evaluation evidence
KEY RESULTS
PO4AO (Core ConceptsPolicyThe rule or model that maps observations or states to actions. Optimization for AO) consistently outperformed standard integrator Control & PlanningControlThe method used to make the robot move the way you want. across varying atmospheric conditions and flux levels on-sky, handling vibration compensation and measurement Data, Distributions & Training IssuesNoiseUnwanted variation or randomness in sensor readings or actuation. robustly with a single hyperparameter set.
WHY DEVELOPERS SHOULD CARE
This is the first real-world Simulation & Sim-to-RealDeploymentPutting the trained system on a real robot. of RL-based adaptive optics Control & PlanningControlThe method used to make the robot move the way you want. on an actual telescope, proving that learned Control & PlanningControlThe method used to make the robot move the way you want. policies can outperform traditional integrator controllers while automatically handling vibrations and Data, Distributions & Training IssuesNoiseUnwanted variation or randomness in sensor readings or actuation.. For roboticists, it validates that Imitation & Reinforcement LearningReinforcement Learning (RL)Teaching a robot through trial and error using rewards. can scale from Simulation & Sim-to-RealSimulationA virtual environment where robots can be trained or tested. to high-frequency real-world Control & PlanningControlThe method used to make the robot move the way you want. tasks with tight Simulation & Sim-to-RealLatencyDelay between input, computation, and action. budgets and changing conditions—without per-condition retuning.
LIMITATIONS
The main limitation to check is whether the claimed behavior holds outside the paper's reported setup. That means testing across different Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. embodiments, scenes, objects, and data distributions.
WHAT COMES NEXT
The practical next step is independent reproduction with clear baselines, ablations, and stress tests. For a developer, the useful follow-up is to map the paper's Imitation & Reinforcement LearningReinforcement Learning (RL)Teaching a robot through trial and error using rewards. assumptions onto a concrete Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. stack, then test the smallest version of the method that could run end to end.