eduzhai > Applied Sciences > Engineering >

Data Driven Control with Learned Dynamics Model-Based versus Model-Free Approach

  • Save

... pages left unread,continue reading

Document pages: 17 pages

Abstract: This paper compares two different types of data-driven control methods,representing model-based and model-free approaches. One is a recently proposedmethod - Deep Koopman Representation for Control (DKRC), which utilizes a deepneural network to map an unknown nonlinear dynamical system to ahigh-dimensional linear system, which allows for employing state-of-the-artcontrol strategy. The other one is a classic model-free control method based onan actor-critic architecture - Deep Deterministic Policy Gradient (DDPG), whichhas been proved to be effective in various dynamical systems. The comparison iscarried out in OpenAI Gym, which provides multiple control environments forbenchmark purposes. Two examples are provided for comparison, i.e., classicInverted Pendulum and Lunar Lander Continuous Control. From the results of theexperiments, we compare these two methods in terms of control strategies andthe effectiveness under various initialization conditions. We also examine thelearned dynamic model from DKRC with the analytical model derived from theEuler-Lagrange Linearization method, which demonstrates the accuracy in thelearned model for unknown dynamics from a data-driven sample-efficientapproach.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×