In modern control applications such as soft robotics or human-robot-interaction, the control design and modelling process becomes very time-consuming or even infeasible due to the complexity of the systems. A solution to overcome these issues is provided by data-driven models which require only a minimum of prior knowledge. However, a general problem of data-driven models is the quantification of the model error which is key for safe application in control. Within the past two decades, Gaussian processes (GPs) have been used increasingly as a data-driven technique due to many beneficial properties such as the bias-variance trade-off and the strong connection to Bayesian mathematics. In contrast to most of the other techniques, GP models provide not only a prediction but also a measure for the uncertainty of the prediction. This powerful property makes them very attractive for many applications in control, e.g., model predictive control, robust control, reinforcement learning and general optimization tasks as the uncertainty measure allows to provide performance and safety guarantees.
The interactive website allows to explore Gaussian process regression for educational purposes. It can be observed how the GP generalizes well even for small data sets and how the selection/optimization of the hyperparameters affects the regression.
The website is written in HTML5 and JavaScript.
The website is publicly available at: https://apps.tbeckers.com/gpr and as Android app: https://play.google.com/store/apps/details?id=com.tbeckers.gpr.twa
By Thomas Beckers