Core Concepts
The author argues that a control-oriented experiment design approach can enhance data-driven controllers by focusing on improving closed-loop performance. This is achieved through stochastic gradient descent and certainty equivalence in linear dynamics.
Abstract
The content discusses an experiment design approach for the linear quadratic regulator, emphasizing control-oriented strategies over traditional methods. It explores offline experiment design settings, system identification steps, and gradient estimation techniques to optimize controller performance.
Key points include:
Proposal of a control-oriented approach for data-driven control.
Comparison with classical A-optimal experiment design.
Focus on linear dynamics and quadratic objectives.
Solution method using stochastic gradient descent.
Demonstration of improved control performance through experimental results.
The study highlights the importance of tailored experiment designs in enhancing controller efficiency and overall system performance.
Stats
"We consider an offline experiment design approach to gathering data where we design a control input to collect data that will most improve the performance of a feedback controller."
"Our method outperforms A-optimal design in terms of improving control performance."
"We show how such a control-oriented approach to experiment design can be carried out for the control of a linear system with uncertain matrix dynamics and a quadratic objective function."