We develop a hybrid controller that drives a nonholonomic vehicle to the source of a radiation-like signal. The control signal depends only on measurements of the signal, which may be corrupted by noise, and internal states of the controller. The control strategy is inspired by a line minimization-based algorithm for unconstrained optimization of nonlinear functions without gradient information. We state the main properties of the algorithm and then discuss its implementation in a hybrid controller that coordinates the optimization algorithm with vehicle steering. We present semiglobal practical convergence and stability results for the closedloop system. These results are illustrated by simulation and experiment.