Pattern Search for Closed-loop Parameter Estimation of Two Wheel Inverted Pendulum System

This paper investigates the pattern search technique for parameters estimation of the Two Wheel Inverted Pendulum (TWIP) system using a grey-box modeling approach. A family pattern search methods like positive basis NP1 active set, basis 2N active set, and Latin hypercube active set are compared against other optimization methods such as Nonlinear least squares and gradient descent. Estimated models are studied under different test scenarios and excitation signals. The parameters obtained with pattern search have the best model matching with real-time measurement based on the cost function. A nonlinear differential equation is obtained and validated for further model-based control.