Augmented Reality Demonstrations for Scalable Robot Imitation Learning

In contrast, Robot Imitation Learning (IL) is a widely used method for training Imitation Learning (IL) aims to empower end-users to teach robots robots to perform manipulation tasks that involve mimicking skills and behaviors through demonstrations, showing promising human demonstrations to acquire skills. However, its practicality results in controlled laboratory environments [3, 4, 11, 13]. However, has been limited due to its requirement that users be trained current methods for collecting demonstrations require users in operating real robot arms to provide demonstrations. This paper to be acquainted with the operation of specific controllers or engage presents an innovative solution: an Augmented Reality (AR)- in contact-based kinesthetic teaching on real robot arms [6, 16, 18], assisted framework for demonstration collection, empowering nonroboticist thereby impeding the widespread application of IL. users to produce demonstrations for robot IL using devices To streamline demonstration collection for non-roboticists, it like the HoloLens 2. Our framework facilitates scalable and is crucial to address two issues: 1) non-expert users typically lack diverse demonstration collection for real-world tasks. We validate understanding of robot arm controllers, and 2) non-roboticists may our approach with experiments on three classical robotics tasks: face limited access to real robot arms due to their high cost and the reach, push, and pick-and-place. The real robot performs each task specialized nature of robot manipulators. While virtual reality (VR) successfully while replaying demonstrations collected via AR. has been used for addressing these challenges [8, 21], it requires