Input‑force‑error convolved Gaussian process for predicting spring‑back pose deviations in satellite deployable arms

The mating of rods and joints in satellite deployable arm is a crucial assembly step that directly affects the accuracy of the final assembly. Upon connection of the rods and joints, the constrained positions induce assembly forces, causing elastic deformation (spring-back) of the rods and resulting in pose deviations after assembly. To address this issue, this paper proposes a method to predict spring-back pose deviations of deployable arm. We introduce a Gaussian process with input-force-error surrogate that predicts the final six-dimensional spring-back pose deviations directly from assembly forces. By pre-evaluating pose deviations in this way, the approach significantly reduces the workload involved in measuring pose deviations, thus improving compensation efficiency. By analytically convolving the Gaussian process kernel with the statistical distribution of Force/Torque (F/T) sensor noise, GPIFE embeds measurement uncertainty directly into the model. Compared with stochastic Kriging, universal Kriging, and multivariate regression, it lowers the translational and rotational root mean square deviations to 0.0329 mm and 5.27 × 10⁻⁴°, delivering improvements of 17% and 4% over the best competing method. Noise-sensitivity experiments spanning a wide range of F/T sensor noise levels further confirm GPIFE’s superior robustness and predictive reliability.