This thesis proposes a novel control architecture for internet-based personal robot systems. The personal robot considered in this thesis is a kind of service robots which can be used for a person's convenience in the house. It has a wireless LAN system for the internet remote control. Users can control the personal robot using a simulator located at local sites. Since the internet time delay is affected by the number of nodes and the internet loads, it is variable and unpredictable so that a large internet delay makes some control inputs distorted. In order to control the personal robot without the influence of the internet delay, a novel internet control architecture is designed. This architecture guarantees that the robot can avoid obstacles, and reduces the path error and the time difference between the virtual robot in a local site and the real robot in a remote site. It consists of a posture estimator based on the posture information from the remote site, a command filter constructed by a command queue and a command generator, a path generator for each sample time and a path-following controller to recover the time difference. The difference between a virtual environment in local sites and a real environment in remote sites can increase a path error between a virtual robot and a real robot. To solve this problem, a virtual environment supervisor with a sensor feedback from the real robot is proposed. The results of simulations and experiments demonstrate the effectiveness and applicability of the proposed internet control architecture.