A nonholonomic under-actuated robot with a bounded control range travels in a 3D workspace, which hosts an unknown domain. The robot has access to a certain direction (typically, vertical) in the space, its own coordinate in this direction, and the horizontal distance to the domain. We present a new navigation law that drives the robot to the desired distance from the domain and then maintains this distance and ensures scan-coverage of the domain boundary within a given range of "altitudes". This law solves the navigation problem via generating the current control as a reflex-like reaction to the current observation and without building any analytical representation or map of the boundary. Mathematically rigorous justification of the proposed guidance approach is provided; its applicability and performance are confirmed by computer simulation tests. (C) 2016 Elsevier Ltd. All rights reserved.