This paper presents a decentralized mobility control algorithm for the formation and maintenance of an optimal cascaded communication chain between a lead sensor-equipped robot and a control station, using a team of robotic vehicles acting as communication relays in an unknown and dynamic RF environment. The gradient-based controller presented uses measurements of the signal-to-noise ratio (SNR) field of neighbor communication links, as opposed to relative position between nodes, as input into a localized performance function. By using the SNR field as input into the control system, the controller is reactive to unexpected and unpredictable changes in the RF environment that is not possible with range-based controllers. Since the operating environment is not known a priori to deployment of a robotic sensor network, an adaptive model-free extremum seeking (ES) algorithm, that uses the motion of the relays to estimate the performance function gradient, is presented to control the motion of 2D nonholonomic vehicles acting as communication relays using the gradient-based controller. Even without specific knowledge of the SNR field, simulations show that the ES decentralized chaining controller using measurements of the SNR field, will drive a team of robotic vehicles to locations that achieve the global objective of maximizing capacity of a cascaded communication chain, even in the presence of an active jamming source.