Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enabling Robots to Identify Missing Steps in Robot Tasks for Guided Learning from Demonstration

Maximilian Diehl, Tathagata Chakraborti, Karinne Ramirez-Amaro

TL;DR

The paper tackles the inefficiency of Learning from Demonstration (LfD) when robots lack knowledge of missing sub-tasks in a task. It introduces guided demonstrations via minimal excuse states $I \Delta I'$ found by combinatorial search, enabling humans to teach only the missing sub-task needed to enable the robot to solve the goal $G_R$ with its existing skills. The approach is operationalized through a VR teaching setup, automatic domain generation from demonstrations, and an automated excuse-generation algorithm based on model reconciliation, with a pilot study showing on average a 61% reduction in demonstration time and a 72% reduction in demonstration size. The findings suggest guided demonstrations can substantially improve LfD efficiency, though challenges like misinterpretation and preference modeling remain for future work, potentially aided by LLMs for richer excuses.

Abstract

Learning from Demonstration (LfD) systems are commonly used to teach robots new tasks by generating a set of skills from user-provided demonstrations. These skills can then be sequenced by planning algorithms to execute complex tasks. However, LfD systems typically require a full demonstration of the entire task, even when parts of it are already known to the robot. This limitation comes from the system's inability to recognize which sub-tasks are already familiar, leading to a repetitive and burdensome demonstration process for users. In this paper, we introduce a new method for guided demonstrations that reduces this burden, by helping the robot to identify which parts of the task it already knows, considering the overall task goal and the robot's existing skills. In particular, through a combinatorial search, the method finds the smallest necessary change in the initial task conditions that allows the robot to solve the task with its current knowledge. This state is referred to as the excuse state. The human demonstrator is then only required to teach how to reach the excuse state (missing sub-task), rather than demonstrating the entire task. Empirical results and a pilot user study show that our method reduces demonstration time by 61% and decreases the size of demonstrations by 72%.

Enabling Robots to Identify Missing Steps in Robot Tasks for Guided Learning from Demonstration

TL;DR

The paper tackles the inefficiency of Learning from Demonstration (LfD) when robots lack knowledge of missing sub-tasks in a task. It introduces guided demonstrations via minimal excuse states found by combinatorial search, enabling humans to teach only the missing sub-task needed to enable the robot to solve the goal with its existing skills. The approach is operationalized through a VR teaching setup, automatic domain generation from demonstrations, and an automated excuse-generation algorithm based on model reconciliation, with a pilot study showing on average a 61% reduction in demonstration time and a 72% reduction in demonstration size. The findings suggest guided demonstrations can substantially improve LfD efficiency, though challenges like misinterpretation and preference modeling remain for future work, potentially aided by LLMs for richer excuses.

Abstract

Learning from Demonstration (LfD) systems are commonly used to teach robots new tasks by generating a set of skills from user-provided demonstrations. These skills can then be sequenced by planning algorithms to execute complex tasks. However, LfD systems typically require a full demonstration of the entire task, even when parts of it are already known to the robot. This limitation comes from the system's inability to recognize which sub-tasks are already familiar, leading to a repetitive and burdensome demonstration process for users. In this paper, we introduce a new method for guided demonstrations that reduces this burden, by helping the robot to identify which parts of the task it already knows, considering the overall task goal and the robot's existing skills. In particular, through a combinatorial search, the method finds the smallest necessary change in the initial task conditions that allows the robot to solve the task with its current knowledge. This state is referred to as the excuse state. The human demonstrator is then only required to teach how to reach the excuse state (missing sub-task), rather than demonstrating the entire task. Empirical results and a pilot user study show that our method reduces demonstration time by 61% and decreases the size of demonstrations by 72%.
Paper Structure (21 sections, 1 figure, 2 tables)

This paper contains 21 sections, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. The Guided Demonstration Task
  4. Measures for Guided Demonstration Effectiveness
  5. Reduction in Demonstration Size
  6. Fraction of Useful Demonstrations
  7. Misdirected Guidance
  8. Plan Solvability after Demonstration
  9. Caveats in Using Excuses as Guidance
  10. Redundancy of the Excuse
  11. Preferred Excuse for Guidance
  12. Used Demonstration and Excuse Generation Methods
  13. Automatic Domain Generation from Demonstrations
  14. Automated Excuse Generation Algorithm
  15. User Study
  16. ...and 6 more sections

Figures (1)

  • Figure 1: The Kitchen Domain (\ref{['fig:drawer_domain']}-\ref{['fig:drawer_chair_domain_ego']}) represents the class of HomeWorld domains wisspeintner2009robocup where the robot has to store a plate in a drawer, but initially does not know how to open a drawer (\ref{['fig:drawer_domain']}-\ref{['fig:drawer_domain_ego']}) and how to unblock the drawer from a chair (\ref{['fig:drawer_chair_domain']}-\ref{['fig:drawer_chair_domain_ego']}). Instead of having to demonstrate the full tasks, our proposed approach of guided demonstrations facilitates the teaching process, by instructing the human to only teach how to reach the automatically generated excuse states where the drawer is open (\ref{['fig:drawer_domain']}) and the drawer is clear (\ref{['fig:drawer_chair_domain']}). We directly communicate the obtained excuse states by displaying its symbolic state description in the VR environment used for the teaching process (Open PinkDrawer - Fig. \ref{['fig:drawer_domain']} and Clear PinkDrawer - Fig. \ref{['fig:drawer_chair_domain']}) as seen at the top of the respective images.

Theorems & Definitions (4)

  • Definition 1
  • Definition 2
  • Definition 3
  • Definition 4