In constrained space, the wrist force sensor of reconfigurable modular robot is expensive and easily affected by environment, it also increased the complexity of the robot used in software and hardware design, so in order to solve this problem, the force/position control method for reconfigurable modular robot based on the soft sensor is proposed. First, the model of the reconfigurable modular robot system was decomposed in all directions. The model in each direction was considered as a subsystem, and the adaptive RBF neural network was used to estimate the uncertainty and coupling relation among subsystems. Then, in the end was not installed the sensor's case, the adaptive RBF neural network was used to estimate the contact force, and the weights of the adaptive RBF neural network, adaptive laws of centers and widths of radial basis function were derived. Finally, the simulation results of 2 kinds of reconfigurable modular robots with different degrees of freedom (DOF) verify the effectiveness of the proposed method.