Spatial Sampling for Alignment of Robot Demonstrated Trajectories in Upper Limb Rehabilitation Tasks

In robotics, Learning by Demonstration (LbD) aims to transfer skills to robots by leveraging multiple demonstrations of the same task. These demonstrations are stored in a library and processed to extract a consistent skill representation, typically requiring temporal alignment using techniques like Dynamic Time Warping (DTW). In this article, we propose a novel Spatial Sampling (SS) algorithm tailored for robot trajectories, which enables time-agnostic alignment by providing an arc-length parametrization of the input trajectories. This method eliminates the need for temporal alignment and enhances skill representation. We demonstrate the effectiveness of SS in an upper-limb rehabilitation case study, introducing a new human-robot interaction architecture.