<p>Learning from Demonstration, in which robots learn our human complicated operational jobs, is simpler than conventional script authoring for task execution, and it is often desired for robots to swiftly adjust to changes in the environment via human demonstrations. For the robot trajectory learning issue, we propose a framework for recovery learning of robot operational skills based on gaussian process. The mean function of the multi-output Gaussian process is derived from Gaussian mixed regression, which we refer to as the mean-prior multi-output Gaussian process, which takes into account the variability of the demonstration teaching and the adaptability of the recovetry skill. In our experiments, our method is compared with Gaussian mixture regression and kernelized movement primitives in a two-part evaluation. The first part is a simulation experiment for handwritten letter trajectory reproduction, and the second part is a robot performing a pegging experiment. The robot trajectory diagram shows that our method is more consistent with the instructor’s intention in the regression trajectory and can complete the task by crossing obstacles in real experiments.</p>