Adaptive Parameter Estimation of a Steerable Drifter

Steerable drifters are a promising energy-efficient environmental sampling platform for aquatic environments with pronounced flows, such as rivers and lakes or oceans with circulation structures. Due to aging and environmental variability, the dynamics of a drifter are often uncertain, which poses challenges in achieving desired control of the robot. In this paper online parameter estimation is explored for a drifter model that is highly nonlinear. With a gradient descent method, the convergence behavior of the adaptation law is explored for two different flow conditions, a parabolic flow and a uniform flow. The influence of the system input, the rudder angles, is also examined for two cases, fixed and sinusoidally varying. The region and speed of parameter convergence are studied in terms of the initial parameter estimate via simulation, which results in a number of findings. For example, for the parabolic flow setting, the region of convergence is much larger than that for the uniform flow setting, with a generally faster convergence, and varying the input results in faster convergence than holding the input constant. Furthermore, for each flow setting and input, there is a clear region of initial parameter estimates for which convergence is achieved more quickly than other parameter estimates. Convergence time is found to depend mostly on the distance of the initial parameter estimate from a trench in the parameter space. This could be useful for informing initial parameter estimates. The spectral content of the regressor is then examined to gain insight into the adaptive behavior.