Automatic parameter learning for easy instruction of industrial collaborative robots

The manufacturing industry faces challenges in meeting requirements of flexibility, product variability and small batch sizes. Automation of high mix, low volume productions requires faster (re)configuration of manufacturing equipment. These demands are to some extend accommodated by collaborative robots. Certain actions can still be hard or impossible to manually adjust due to inherent process uncertainties. This paper proposes a generic iteratively learning approach based on Bayesian Optimisation to efficiently search for the optimal set of process parameters. The approach takes into account the process uncertainties by iteratively making a statistical founded choice on the next parameter-set to examine only based on the prior binomial outcomes. Moreover, our function estimator uses Wilson Score to make proper estimates on the success probability and the associated uncertain measure of sparsely sampled regions. The function estimator also generalises the experiment outcomes to the neighbour region through kernel smoothing by integrating Kernel Density Estimation. Our approach is applied to a real industrial task with significant process uncertainties, where sufficiently robust process parameters cannot intuitively be chosen. Using our approach, a collaborative robot automatically finds a reliable solution.