This article presents a cross-domain robot (CDR) that experiences drive efficiency degradation when operating on water surfaces, similar to drive faults. Moreover, the CDR mathematical model has uncertain parameters and non-negligible water resistance. To solve these problems, a radial basis function neural network (RBFNN)-based active fault-tolerant control (AFTC) algorithm is proposed for the robot both on land and water surfaces. The proposed algorithm consists of a fast non-singular terminal sliding mode controller (NTSMC) and an RBFNN. The RBFNN is used to estimate the impact of drive faults, water resistance, and model parameter uncertainty on the robot and the output value compensates the controller. Additionally, an anti-input saturation control algorithm is designed to prevent driver saturation. To optimize the controller parameters, a human decision search algorithm (HDSA) is proposed, which mimics the decision-making process of a crowd. Simulation results demonstrate the effectiveness of the proposed control methods.