Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Framework for Learning and Reusing Robotic Skills

Brendan Hertel, Nhu Tran, Meriem Elkoudi, Reza Azadeh

TL;DR

The paper tackles learning and reusing robotic skills by proposing a library of motion primitives learned from Learning from Demonstration (LfD). It introduces a multimodal probabilistic segmentation to decompose demonstrations into primitives, followed by elastic maps-based clustering to automatically discover the number of primitive types, and a trajectory-editing approach (LTE) to adapt primitives to new environments. The framework is validated through simulations and real-robot demonstrations, showing that primitive discovery and organization are feasible and that primitives can be reused across contexts. The work advances intuitive skill learning and reuse, with potential impact on building scalable robotic skill libraries and enabling higher-level planning from learned primitives.

Abstract

In this paper, we present our work in progress towards creating a library of motion primitives. This library facilitates easier and more intuitive learning and reusing of robotic skills. Users can teach robots complex skills through Learning from Demonstration, which is automatically segmented into primitives and stored in clusters of similar skills. We propose a novel multimodal segmentation method as well as a novel trajectory clustering method. Then, when needed for reuse, we transform primitives into new environments using trajectory editing. We present simulated results for our framework with demonstrations taken on real-world robots.

A Framework for Learning and Reusing Robotic Skills

TL;DR

The paper tackles learning and reusing robotic skills by proposing a library of motion primitives learned from Learning from Demonstration (LfD). It introduces a multimodal probabilistic segmentation to decompose demonstrations into primitives, followed by elastic maps-based clustering to automatically discover the number of primitive types, and a trajectory-editing approach (LTE) to adapt primitives to new environments. The framework is validated through simulations and real-robot demonstrations, showing that primitive discovery and organization are feasible and that primitives can be reused across contexts. The work advances intuitive skill learning and reuse, with potential impact on building scalable robotic skill libraries and enabling higher-level planning from learned primitives.

Abstract

In this paper, we present our work in progress towards creating a library of motion primitives. This library facilitates easier and more intuitive learning and reusing of robotic skills. Users can teach robots complex skills through Learning from Demonstration, which is automatically segmented into primitives and stored in clusters of similar skills. We propose a novel multimodal segmentation method as well as a novel trajectory clustering method. Then, when needed for reuse, we transform primitives into new environments using trajectory editing. We present simulated results for our framework with demonstrations taken on real-world robots.
Paper Structure (9 sections, 3 equations, 4 figures, 1 algorithm)

This paper contains 9 sections, 3 equations, 4 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Segmentation
  4. Clustering
  5. Selection of Reusable Primitives
  6. Experiments
  7. Segmentation
  8. Clustering
  9. Future Work

Figures (4)

  • Figure 1: The proposed framework for creating a library of motion primitives.
  • Figure 2: The proposed probabilistic multimodal segmentation method.
  • Figure 3: Segmenting a 2D handwritten R shape and a real-world demonstration of pressing multiple buttons. a) The jerk profile of the shape, with changepoint segments shown in different colors. b) the probabilities of keypoints in time. c) the segmented shape with different segments shown in different colors. d) The demonstration for pressing multiple buttons on a real-world UR5e robot. e) Multimodal segmentation performed on a real-world demonstration of pressing multiple buttons.
  • Figure 4: Results of elastic clustering and agglomerative clustering applied to the RAIL Dataset rana2020benchmark. Elastic clustering provides better results as less demonstrations are wrongly clustered.