We present an evolution of an agent, remotely operating a fast running scale model of a car. The agent perceives the environment from overhead video camera and conveys its actions via radio control transmitter. In order to cope with the video feed latency we propose an anticipatory modeling in which the agent considers its actions based on the anticipated state of the car. The agent is first evolved offline on a software simulator and then adapted online to the real world. During the online evolution, the lap times improve to the values much close to the values obtained from the offline evolution. An online evolutionary optimization of the avoidance of a small static obstacle with a priori known properties results in lap times that are virtually the same as the best lap times achieved on the same track without obstacles. This work can be viewed as a step towards the automated design of controllers of remotely operated vehicles capable to find an optimal solution to various tasks in a priori known environments.