The task of this paper is to find lost or frozen people in the wood. That takes accurate exploration of a large space with a minimum time duration. This work is dedicated to the architecture part of the assigned task. We give an architecture for robot-finder capable to find a human being in the wood or in the snow area. For us, the robot is blending of two elements, which we can develop independently. That are a wheeled platform and an operating module. In this task, we look at the second one. During that way we assume that the first one is developed, therefore the robot is driving upon the airbag or wheeled platform. Our solution to this task is architecture and algorithm. These two are made for and directed to learn robot follow the map and detect human being alongside. We use computer vision, neural network and GPS technologies. In the end, we have a theoretical basis for developing robot-finder.