Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Online Learning of Human Constraints from Feedback in Shared Autonomy

Shibei Zhu, Tran Nguyen Le, Samuel Kaski, Ville Kyrki

TL;DR

This work proposes an augmentative assistant agent capable of learning and adapting to human physical constraints, aligning its actions with the ergonomic preferences and limitations of the human operator.

Abstract

Real-time collaboration with humans poses challenges due to the different behavior patterns of humans resulting from diverse physical constraints. Existing works typically focus on learning safety constraints for collaboration, or how to divide and distribute the subtasks between the participating agents to carry out the main task. In contrast, we propose to learn a human constraints model that, in addition, considers the diverse behaviors of different human operators. We consider a type of collaboration in a shared-autonomy fashion, where both a human operator and an assistive robot act simultaneously in the same task space that affects each other's actions. The task of the assistive agent is to augment the skill of humans to perform a shared task by supporting humans as much as possible, both in terms of reducing the workload and minimizing the discomfort for the human operator. Therefore, we propose an augmentative assistant agent capable of learning and adapting to human physical constraints, aligning its actions with the ergonomic preferences and limitations of the human operator.

Online Learning of Human Constraints from Feedback in Shared Autonomy

TL;DR

This work proposes an augmentative assistant agent capable of learning and adapting to human physical constraints, aligning its actions with the ergonomic preferences and limitations of the human operator.

Abstract

Real-time collaboration with humans poses challenges due to the different behavior patterns of humans resulting from diverse physical constraints. Existing works typically focus on learning safety constraints for collaboration, or how to divide and distribute the subtasks between the participating agents to carry out the main task. In contrast, we propose to learn a human constraints model that, in addition, considers the diverse behaviors of different human operators. We consider a type of collaboration in a shared-autonomy fashion, where both a human operator and an assistive robot act simultaneously in the same task space that affects each other's actions. The task of the assistive agent is to augment the skill of humans to perform a shared task by supporting humans as much as possible, both in terms of reducing the workload and minimizing the discomfort for the human operator. Therefore, we propose an augmentative assistant agent capable of learning and adapting to human physical constraints, aligning its actions with the ergonomic preferences and limitations of the human operator.
Paper Structure (10 sections, 5 equations, 5 figures)

This paper contains 10 sections, 5 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Problem formulation
  3. Learning a trust region from real-time feedback with human constraint model
  4. Human's feedback as indirect labels for the negative and positive samples.
  5. Potential use cases
  6. Co-transportation.
  7. Rehabilitation robots.
  8. Exoskeleton robots.
  9. Experiments
  10. Conclusion

Figures (5)

  • Figure 1: Co-transportation task where both human and robot operate on the same object. Left: assistive agent that considers human physical constraints. Right: assistive agent without considering human's physical constraints, and human operator tries to retake control by pulling back the object.
  • Figure 2: Constraints in the task space where different agents have their constraints in the action space. Given the behavior spaces of humans and robots constrained by their respective physical constraints, the colored areas represent the trust region that defines the plausible set of joint behavior that satisfies all the constraints. We represent the lower bound of the trust region as the constraints that define the upper bound of the human behavior space.
  • Figure 3: Given the external feedback received in real-time from a human, we develop a human constraint model to define a lower bound of the trust region where the joint actions of both the human and the assistive agent satisfy their constraints.
  • Figure 4: Robot assists human in co-transportation task (Left) and rehabilitation (Right).
  • Figure 5: Real human feedback from 2 individuals during the collaborative tasks, represented by their means and standard deviations. The force feedback is six-dimensional vector that is represented by each plot. The x-axis represents the time, and the y-axis the corresponding value reading. 5 runs are used to generate these plots.