NEW RELEASE: A Case Study in Robotics

The article "Minimum Cost Averaging for Multivariate Time Series Using Constrained Dynamic Time Warping: A Case Study in Robotics" was published on the 1st August 2023 by Itziar Cabanes, Miguel Prada, Anthony Remazeilles, Irati Rasines.

In this short Interview, Irati Rasines from TECNALIA gives a quick overview about the main topics.

What is the article about? What are the main results it presents?

In robotics, the learning by demonstration consists in giving to the robot one or several illustrations of the behavior to reproduce. The data collected is used by the robot to learn the desired action. Using several demonstrations raises several issues to handle: the velocity may be different across demonstrations, the execution may be slightly different… we propose an original approach to solve this case.

This article presents a new multiple multivariate time series averaging technique based on Constrained Dynamic Time Warping. MCA-CDTW is a task-agnostic approach that after selecting a reference curve, transforms the rest of the demonstrations in the set to obtain new curves that are time-aligned with the reference. This technique provides smooth mean curves even when there are large deviations between the demonstrations in the set, and still the complexity of the proposed algorithm is significantly reduced compared to other averaging techniques from the literature.

What/who were the study objects/participants and why were they selected?

The new algorithm MCA-CDTW is tested and compared on two different databases: a literature database where humans move a robotic arm with kinaesthetic teaching, and a set of recordings of a teleoperated robotic arm performing laboratory manipulation.

Both databases contain several demonstrations of a given motion. We prove that our approach is able to extract a nominal curve to learn, and compare the smoothness of this curve with other techniques from the literature.

 How is the paper connected to the TraceBot Project?

As MCA-CDTW creates transformed curves that are time-aligned, not only the nominal behavior is obtained but also the variability around the average curve also at no additional cost. In a robotic scenario, such as the one in Tracebot, such capability is an added value as it provides information about where the robot should be stiffer following the human demonstrations.

The second database used in the experimentation is directly connected to TraceBot, as it deals with the manipulation of the canister and insertion into the pump tray. We are investigating the possibility of using teleoperation to teach the robot specific motions in the sterility testing process. We consider that using several demonstrations of a given motion enables obtaining a better description of the task, but it raises the question of how to combine these multiple illustrations, which is the topic of the article.


Graphical abstract of the paper: A new task agnostic multivariate time series averaging technique that provides smooth average curves. This can be essential in fields such as robotics.

Go back