eduzhai > Applied Sciences > Engineering >

Learning Event-triggered Control from Data through Joint Optimization

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 18 pages

Abstract: We present a framework for model-free learning of event-triggered controlstrategies. Event-triggered methods aim to achieve high control performancewhile only closing the feedback loop when needed. This enables resourcesavings, e.g., network bandwidth if control commands are sent via communicationnetworks, as in networked control systems. Event-triggered controllers consistof a communication policy, determining when to communicate, and a controlpolicy, deciding what to communicate. It is essential to jointly optimize thetwo policies since individual optimization does not necessarily yield theoverall optimal solution. To address this need for joint optimization, wepropose a novel algorithm based on hierarchical reinforcement learning. Theresulting algorithm is shown to accomplish high-performance control in linewith resource savings and scales seamlessly to nonlinear and high-dimensionalsystems. The method s applicability to real-world scenarios is demonstratedthrough experiments on a six degrees of freedom real-time controlledmanipulator. Further, we propose an approach towards evaluating the stabilityof the learned neural network policies.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×