<p>Wire-driven flexible arms exhibit characteristics of high redundancy in degrees of freedom and nonlinear motion, which result in deformations and motion trajectories that are difficult to accurately calculate during force application and movement. Traditional rigid robotic arm motion models are no longer applicable, and more realistic computational models or methods are required. To address this, this paper benchmarks experimental data from the motion process of flexible robotic arms, collects data on the deflection angles of connecting plates, driving wire tensions, and end-point coordinates using multiple sensors, and correlates this data with actual poses calculated from multi-angle images. A mathematical model is established by computing the motion poses through multi-sensor data fusion. To further ensure model accuracy, a random forest model algorithm is used to perform machine learning on the collected state information and the spatial poses of the flexible arm. Experiments show that analyzing sensor data through intelligent algorithms can accurately predict the three-dimensional coordinates of the flexible arm’s end-point and optimize the kinematics model of the flexible arm. The average error between calculations and experimental results is less than 1.2% for C-shaped poses and less than 6.2% for S-shaped poses. In summary, the combination of multi-sensor data fusion and artificial intelligence algorithms can effectively improve the accuracy of judging and predicting the end-point position of the flexible arm as well as the overall pose of the flexible arm.</p>