Adaptive optics control with multi-agent model-free reinforcement learning.

We present a novel formulation of closed-loop adaptive optics (AO) control as a multi-agent reinforcement learning (MARL) problem in which the controller is able to learn a non-linear policy and does not need a priori information on the dynamics of the atmosphere. We identify the different challenges of applying a reinforcement learning (RL) method to AO and, to solve them, propose the combination of model-free MARL for control with an autoencoder neural network to mitigate the effect of noise. Moreover, we extend current existing methods of error budget analysis to include a RL controller. The experimental results for an 8m telescope equipped with a 40x40 Shack-Hartmann system show a significant increase in performance over the integrator baseline and comparable performance to a model-based predictive approach, a linear quadratic Gaussian controller with perfect knowledge of atmospheric conditions. Finally, the error budget analysis provides evidence that the RL controller is partially compensating for bandwidth error and is helping to mitigate the propagation of aliasing.

Cooperative binding of the retinal rod G-protein, transducin, to light-activated rhodopsin.

Direct measurements of the binding between light-activated rhodopsin (Rho*) and transducin, the retinal rod G-protein, revealed a strongly cooperative interaction. Cooperativity was assessed by measuring the association of 125I-labeled transducin (Gt) to Rho* in urea-stripped rod outer segment membranes at equilibrium. Analysis of 125I-Gt binding curves gave a Hill coefficient of 1.8. These data were consistent with a two-site model in which binding of the first 125I-Gt to Rho* increased the binding of the second 125I-Gt approximately 40-fold (Kd values were 80 +/- 30 and 1.9 +/- 0.7 nM, respectively). The effects of GDP on the binding were also investigated. GDP decreased the affinity between Rho* and Gt approximately 100-fold but did not decrease the degree of cooperativity. Binding curves of 125I-Gt in the presence of 1 mM GDP showed a Hill coefficient of 1.9. The data were also consistent with a two-binding site model in which binding of the first 125I-Gt increased the binding of the second 125I-Gt approximately 70-fold (Kd values were 13.7 +/- 5.4 and 0.20 +/- 0.08 microM, respectively). The Gt alpha subunit in the absence of Gt beta gamma also bound Rho* in a cooperative manner. These data implicate a role for the cooperative association of Rho* and Gt in the light activation cascade of retinal rods.