Modeling complex robotic assembly process using Gaussian Process Regression

In the high precision robotic assembly processes, the process parameters have to be tuned in order to adapt to variations and satisfy the performance requirements. For complex systems and processes operating in a stochastic environment such as the assembly processes, experiments and evaluations could be costly and low efficiency because of resource utilization, energy consumption, and dedicated labor. In order to improve the assembly process performance, we investigate the modeling problem for robotic assembly processes. Gaussian Process Regression, a non-parametric modeling technique, is chosen to model the relationship between the assembly process parameters and performance. The main challenge in implementing Gaussian Process Regression is to find suitable covariance functions which can minimize the modeling errors. Therefore we investigated different combinations of basic covariance functions and implemented them to explore the most suitable covariance function for an assembly process. The performance of the built models is compared and the covariance functions with the best performance are identified. An off-line modeling algorithm is appropriately developed using the identified covariance function. The effectiveness and accuracy of the proposed algorithm are further demonstrated by experiments, which were performed using a robotic valve body assembly process.