The thesis deals with the design of adaptive controllers in the form of function approximators. The adaptability of the controller increases the control performance under uncertain model parameters. The approach in the thesis is based on optimized trajectories as a data basis, together with a new approach called disturbed oracle imitation that is required to train a controller with hidden states, e.g., a recurrent neural network. The approach is analyzed empirically using a simulation example and applied to the physical system of a mobile inverted pendulum.
«
The thesis deals with the design of adaptive controllers in the form of function approximators. The adaptability of the controller increases the control performance under uncertain model parameters. The approach in the thesis is based on optimized trajectories as a data basis, together with a new approach called disturbed oracle imitation that is required to train a controller with hidden states, e.g., a recurrent neural network. The approach is analyzed empirically using a simulation example an...
»