Contact State Clustering Analysis Based on Multivariate Time Series

Aiming at the classification problem of the contact state in the robot peg-in-hole task, this paper proposes a clustering method based on the multivariate time series. This method uses a deep temporal clustering network to encode the contact state variables in the assembly process, and then a complexity-invariant distance measure is used to classify the time series fragments. This method avoids the quasi-static analysis of the contact process and thus has a certain generality in practice. And the use of time series is beneficial to extract the time-related characteristics of contact state variables, which can make the clustering results more robust. The experimental results are consistent with expectations, indicating the theoretical correctness and effectiveness of the proposed algorithm.