Extended Radial Basis Function Controller for Reinforcement Learning

  • 2020-09-12 20:56:48
  • Nicholas Capel, Naifu Zhang
  • 0


There have been attempts in model-based reinforcement learning to exploit apriori knowledge about the structure of the system. This paper introduces theextended radial basis function (RBF) controller design. In addition totraditional RBF controllers, our controller comprises of an engineered linearcontroller inside an operating region. We show that the learnt extended RBFcontroller takes on the desirable characteristics of both the linear andnon-linear controller models. The extended controller is shown to retain theability for universal function approximation of the non-linear RBF functions.At the same time, it demonstrates desirable stability criteria on par with thelinear controller. Learning has been done in a probabilistic inferenceframework (PILCO), but could generalise to other reinforcement learningframeworks. Experimental results from the Swing-up pendulum, Cartpole, andMountain car environments are reported.


Quick Read (beta)

loading the full paper ...