Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Investigating the Benefits of Nonlinear Action Maps in Data-Driven Teleoperation

Michael Przystupa, Gauthier Gidel, Matthew E. Taylor, Martin Jagersand, Justus Piater, Samuele Tosatto

TL;DR

It is found that nonlinear odd functions behave linearly for most of the control space, suggesting architecture structure improvements are not the primary factor in data-driven teleoperation.

Abstract

As robots become more common for both able-bodied individuals and those living with a disability, it is increasingly important that lay people be able to drive multi-degree-of-freedom platforms with low-dimensional controllers. One approach is to use state-conditioned action mapping methods to learn mappings between low-dimensional controllers and high DOF manipulators -- prior research suggests these mappings can simplify the teleoperation experience for users. Recent works suggest that neural networks predicting a local linear function are superior to the typical end-to-end multi-layer perceptrons because they allow users to more easily undo actions, providing more control over the system. However, local linear models assume actions exist on a linear subspace and may not capture nuanced actions in training data. We observe that the benefit of these mappings is being an odd function concerning user actions, and propose end-to-end nonlinear action maps which achieve this property. Unfortunately, our experiments show that such modifications offer minimal advantages over previous solutions. We find that nonlinear odd functions behave linearly for most of the control space, suggesting architecture structure improvements are not the primary factor in data-driven teleoperation. Our results suggest other avenues, such as data augmentation techniques and analysis of human behavior, are necessary for action maps to become practical in real-world applications, such as in assistive robotics to improve the quality of life of people living with w disability.

Investigating the Benefits of Nonlinear Action Maps in Data-Driven Teleoperation

TL;DR

It is found that nonlinear odd functions behave linearly for most of the control space, suggesting architecture structure improvements are not the primary factor in data-driven teleoperation.

Abstract

As robots become more common for both able-bodied individuals and those living with a disability, it is increasingly important that lay people be able to drive multi-degree-of-freedom platforms with low-dimensional controllers. One approach is to use state-conditioned action mapping methods to learn mappings between low-dimensional controllers and high DOF manipulators -- prior research suggests these mappings can simplify the teleoperation experience for users. Recent works suggest that neural networks predicting a local linear function are superior to the typical end-to-end multi-layer perceptrons because they allow users to more easily undo actions, providing more control over the system. However, local linear models assume actions exist on a linear subspace and may not capture nuanced actions in training data. We observe that the benefit of these mappings is being an odd function concerning user actions, and propose end-to-end nonlinear action maps which achieve this property. Unfortunately, our experiments show that such modifications offer minimal advantages over previous solutions. We find that nonlinear odd functions behave linearly for most of the control space, suggesting architecture structure improvements are not the primary factor in data-driven teleoperation. Our results suggest other avenues, such as data augmentation techniques and analysis of human behavior, are necessary for action maps to become practical in real-world applications, such as in assistive robotics to improve the quality of life of people living with w disability.

Paper Structure

This paper contains 27 sections, 4 theorems, 19 equations, 19 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. Dynamical Control Systems
  4. Action Mapping Methods
  5. Soft Reversibility
  6. Nonlinear State Action Maps
  7. Regularizing Teleoperation Behavior
  8. State Conditioned Nonlinear Maps
  9. Tensor Layers
  10. Architecture Constraints
  11. Limitations of Nonlinearity
  12. Experiments
  13. Action Map Analysis
  14. Model Regression Quality
  15. Simulated Teleoperation
  16. ...and 12 more sections

Key Result

Theorem 1

In the interval $0 \leq t \leq 1$ following $f$ in Equation eqn:dynamicalsystem, and assuming that $a(t+1) = -a(t-1)$, we have $x(2- t) = x(t)$.

Figures (19)

  • Figure 1: User teleoperating robot with data driven interface.
  • Figure 2: Diagram of neural architecture for action maps.
  • Figure 3: User study environment for data-driven teleoperation experiments. Users had to pick up the cup on the blue box, dispense it's contents into the red container and finally dispense the cup in the green box on the opposite end of the testing environment.
  • Figure 4: Difference between prediction and linearization as action as magnitude increases.
  • Figure 6: Nasa Workload Index between systems during user study. Between all system used by participants we did not find any statistical significants with p-value 0.05 by the Posthoc Dunn mean test.
  • ...and 14 more figures

Theorems & Definitions (8)

  • Theorem 1: Trajectory Reversibility
  • proof
  • Lemma 1
  • proof
  • Theorem 2: Trajectory Soft Reversibility
  • proof
  • Corollary 2.1: Residual Inversing Actions
  • proof