Parameter Estimation using Hybrid Learning by Ryan Johnson

Ryan Johnson

UC Santa Cruz | Nordstrom Stores Endowment Fund

Ryan Johnson’s research in the Hybrid Systems Lab at the University of California, Santa Cruz focuses on parameter estimation and systems identification for hybrid dynamical systems, with applications in power systems, robotics, and aerospace. His recent work proposes a hybrid gradient descent algorithm that achieves finite-time parameter estimation for a class of hybrid systems. The stability properties of the proposed algorithm are established, and its effectiveness is demonstrated using a DC-DC boost converter circuit.


We propose a hybrid gradient descent parameter estimation algorithm for a class of hybrid systems. The algorithm achieves finite-time convergence of the parameter estimate without the need for a persistently exciting regressor. The effectiveness of the proposed estimator is demonstrated using a DC-DC boost converter model. The estimator is interconnected with an adaptive “uniting” control algorithm, which combines global and local control schemes. The global control scheme induces global attractivity of a neighborhood of the desired output voltage value, and the local scheme maintains industry-standard PWM behavior during steady state. The well-posedness and stability properties of the resulting closed-loop hybrid system are established. In particular, the closed-loop hybrid system guarantees global asymptotic stability of a neighborhood of a desired set-point even under uncertainty in the converter input voltage and load resistance.


11 + 11 =