Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Using Machine Teaching to Boost Novices' Robot Teaching Skill

Yuqing Zhu, Endong Sun, Matthew Howard

TL;DR

This work addresses novices' difficulty in teaching robots via learning-from-demonstration by introducing a machine-teaching (MT) based framework that provides guidance to novice teachers. The learner model is linear in features with ridge regression, and teaching risk is minimised under an MT bi-level formulation, with guidance delivered as post-demonstration visual feedback. Through simulated and physical robot experiments, the approach yields substantial improvements in teaching accuracy that persist after training and transfer to new skills, demonstrating both retention and generalisation. The results suggest a scalable pathway to train non-experts to teach robotic systems effectively, potentially reducing the time and data required to deploy new robotic capabilities in real settings.

Abstract

Recent evidence has shown that, contrary to expectations, it is difficult for users, especially novices, to teach robots tasks through LfD. This paper introduces a framework that leverages MT algorithms to train novices to become better teachers of robots, and verifies whether such teaching ability is retained beyond the period of training and generalises such that novices teach robots more effectively, even for skills for which training has not been received. A between-subjects study is reported, in which novice teachers are asked to teach simple motor skills to a robot. The results demonstrate that subjects that receive training show average 78.83% improvement in teaching ability (as measured by accuracy of the skill learnt by the robot), and average 63.69% improvement in the teaching of new skills not included as part of the training.

Using Machine Teaching to Boost Novices' Robot Teaching Skill

TL;DR

This work addresses novices' difficulty in teaching robots via learning-from-demonstration by introducing a machine-teaching (MT) based framework that provides guidance to novice teachers. The learner model is linear in features with ridge regression, and teaching risk is minimised under an MT bi-level formulation, with guidance delivered as post-demonstration visual feedback. Through simulated and physical robot experiments, the approach yields substantial improvements in teaching accuracy that persist after training and transfer to new skills, demonstrating both retention and generalisation. The results suggest a scalable pathway to train non-experts to teach robotic systems effectively, potentially reducing the time and data required to deploy new robotic capabilities in real settings.

Abstract

Recent evidence has shown that, contrary to expectations, it is difficult for users, especially novices, to teach robots tasks through LfD. This paper introduces a framework that leverages MT algorithms to train novices to become better teachers of robots, and verifies whether such teaching ability is retained beyond the period of training and generalises such that novices teach robots more effectively, even for skills for which training has not been received. A between-subjects study is reported, in which novice teachers are asked to teach simple motor skills to a robot. The results demonstrate that subjects that receive training show average 78.83% improvement in teaching ability (as measured by accuracy of the skill learnt by the robot), and average 63.69% improvement in the teaching of new skills not included as part of the training.
Paper Structure (27 sections, 15 equations, 5 figures)

This paper contains 27 sections, 15 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Problem Definition
  3. Materials & Methods
  4. Formulation
  5. Modelling the Learner
  6. Teaching Effort
  7. Teaching Risk
  8. -based Training
  9. Selection of States
  10. Selection of Actions
  11. Guidance Design
  12. Evaluation
  13. Experimental Protocol
  14. Simulated Robot
  15. Set up
  16. ...and 12 more sections

Figures (5)

  • Figure 1: Framework overview. The proposed approach allows human teachers to be trained to teach robot dynamic motor skills using MT.
  • Figure 2: (a) Simulation Platform User Interface: The interface displays a two-axis robotic arm fixed at the origin. The yellow marker and the blue arrow represent the robot's position and velocity, respectively. The dark green arrow shows the action applied by the user, and the light green arrows represent the -based guidance. The progress bar shows the teaching effort so far. (b) Physical Experiment Setup (Top View): The uArm includes a display screen for visual guidance. The black dot shows the current state, and the red dot marks the target. The light green arrow shows the user’s action, and the dark green arrow represents -based guidance. Auditory guidance for teaching effort progress is provided via a speaker. (c) Physical Experiment with Participant: A four-point seat belt ensures the participant maintains a consistent viewpoint.
  • Figure 3: Simulation Experiment Results. (a) indicates the change in the mean $\mathnormal{E_{\ell_2}}$ for the two groups as the experiment progresses. (b)-(d) show the change in $\mathnormal{E_{\ell_2}}$ between \ref{['i:phase1']} and \ref{['i:phase3']}E8, \ref{['i:phase1']} and \ref{['i:phase4']}, \ref{['i:phase2']} and \ref{['i:phase5']}, respectively.
  • Figure 4: Trajectory Comparison. Trajectories produced by representative participants A (from the target group) and B (from the control group). (a) Trajectories from \ref{['i:phase3']}; (b) Trajectories from \ref{['i:phase1']} and \ref{['i:phase4']}; (c) Trajectories from \ref{['i:phase2']} and \ref{['i:phase5']}; (d) Zoomed-in view of participant A's trajectory from \ref{['i:phase3']}.
  • Figure 5: Simulation Experiment Results. (a)-(c) show the change in $\mathnormal{E_{\ell_2}}$ between \ref{['i:phase1']}E1 and \ref{['i:phase3']}E8, \ref{['i:phase1']}E1 and \ref{['i:phase4']}E1, \ref{['i:phase2']}E1 and \ref{['i:phase5']}E1, respectively.