The searching for a location-unknown radio transmitter is a challenging task for autonomous robot. We propose an adaptive searching algorithm named theseus gradient guide (TGG) which is designed for solving the searching problem in indoor environments using received signal strength (RSS). While the RSS gradient serves as the main guide, the robot prefers to move to the places which have never been traveled. Thus the robot will not get stuck in the local maxima. Moreover, unlike the commonly used random kick strategy the TGG drives the robot escaping the local maxima with low cost in terms of travel distance. Meanwhile, TGG is not sensitive to motion errors. Simulation results show that the searches using TGG cost much less compared with those using other gradient based methods in our testing indoor environment. Guided by TGG, the robot can successfully reach the location-unknown radio transmitter with a ratio over 97% when the standard deviation of motion error is up to 20% of the step length.