We compared four algorithms for controlling a MEMS deformable mirror of an adaptive optics (AO) scanning laser ophthalmoscope. Interferometer measurements of the static nonlinear response of the deformable mirror were used to form an equivalent linear model of the AO system so that the classic integrator plus wavefront reconstructor type controller can be implemented. The algorithms differ only in the design of the wavefront reconstructor. The comparisons were made for two eyes (two individuals) via a series of imaging sessions. All four controllers performed similarly according to estimated residual wavefront error not reflecting the actual image quality observed. A metric based on mean image intensity did consistently reflect the qualitative observations of retinal image quality. Based on this metric, the controller most effective for suppressing the least significant modes of the deformable mirror performed the best.

In this paper, an iterative learning controller (ILC) that uses partial but most pertinent information in the error signal from previous cycles is employed for precision control of a wafer stage. Typically, ILC schemes use the error signal from the previous cycle for updating the control input. This error contains both repetitive and nonrepetitive components. The nonrepetitive components of the error cause degradation of performance of the ILC scheme. Based on structural information about the plant and the disturbances, we can determine some basis functions along which the repetitive error is concentrated. This information is extracted by projecting the error signal onto the subspace spanned by these basis functions. The projected error signal is then used in the ILC update law. Stability and convergence conditions are presented for this projection-based ILC update law. The proposed idea is motivated by precision control of a wafer stage. For a constant velocity scan by the wafer stage, the major sources of repetitive error are found to be phase-mismatch and force ripple. These effects are mathematically modeled to obtain the subspace spanned by them. The projection-based ILC scheme using this subspace is then implemented on a prototype one DOF stage and its performance is compared to the standard ILC scheme that uses a frequency-domain filtering to remove nonrepetitive components of the error.

In this work, a contract-based reasoning approach is developed for obstacle avoidance in unmanned aerial vehicles (UAV's) under evolving subsystem performance. This approach is built on an assume-guarantee framework, where each subsystem (guidance, navigation, control and the environment) assumes a certain level of performance from other subsystems and in turn provides a guarantee of its own performance. The assume-guarantee construct then assures the performance of the overall system (in this case, safe obstacle avoidance). The implementation of the assume-guarantee framework is done through a set of contracts that are encoded into the guidance subsystem, in the form of a set of inequality constraints in the trajectory planner. The inequalities encode the relationships between subsystem performance and operational limits that ensure safe and robust operation as the performance of the control and navigation subsystems and environment evolve over time. The contract inequalities can be obtained analytically or numerically using an optimization based path planner and UAV simulation. The methodology is evaluated in the context of head-on obstacle avoidance, where the contracts are constructed in terms of (1) minimum obstacle detection range, (2) expected obstacle size, (3) maximum allowed cruise velocity, (4) maximum allowable thrust, roll and pitch angles, and (5) inner-loop tracking performance. Numerical and analytical generation of these contracts for this scenario is demonstrated. Finally, in-flight contract enforcement is illustrated for typical scenarios.

No abstract