This paper propose a novel approach for modeling the end-to-end time delay dynamics of the Internet using system identification, and use it for controlling real-time Internet-based telerobotic operations. When a single model is used, it needs to adapt to the operating conditions before an appropriate control mechanism can be applied. Slow adaptation may result in large transient errors. As an alternative, we propose to use an adaptive multiple model framework, and determine the best model for the current operating conditions to activate the corresponding controller. We employ multivariable wave prediction method to achieve this objective.