Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Personalised 3D Human Digital Twin with Soft-Body Feet for Walking Simulation

Kum Yew Loke, Sherwin Stephen Chan, Mingyuan Lei, Henry Johan, Bingran Zuo, Wei Tech Ang

TL;DR

This paper proposes to integrate personalised soft-body feet, generated using the motion capture data of real human subjects, into a skeletal model and train it with a walking control policy, and finds that the soft-body feet were able to generate ground reaction force results comparable to real measured data and closely follow joint angle results of the bare skeletal model and the reference motion.

Abstract

With the increasing use of assistive robots in rehabilitation and assisted mobility of human patients, there has been a need for a deeper understanding of human-robot interactions particularly through simulations, allowing an understanding of these interactions in a digital environment. There is an emphasis on accurately modelling personalised 3D human digital twins in these simulations, to glean more insights on human-robot interactions. In this paper, we propose to integrate personalised soft-body feet, generated using the motion capture data of real human subjects, into a skeletal model and train it with a walking control policy. Through evaluation using ground reaction force and joint angle results, the soft-body feet were able to generate ground reaction force results comparable to real measured data and closely follow joint angle results of the bare skeletal model and the reference motion. This presents an interesting avenue to produce a dynamically accurate human model in simulation driven by their own control policy while only seeing kinematic information during training.

Personalised 3D Human Digital Twin with Soft-Body Feet for Walking Simulation

TL;DR

This paper proposes to integrate personalised soft-body feet, generated using the motion capture data of real human subjects, into a skeletal model and train it with a walking control policy, and finds that the soft-body feet were able to generate ground reaction force results comparable to real measured data and closely follow joint angle results of the bare skeletal model and the reference motion.

Abstract

With the increasing use of assistive robots in rehabilitation and assisted mobility of human patients, there has been a need for a deeper understanding of human-robot interactions particularly through simulations, allowing an understanding of these interactions in a digital environment. There is an emphasis on accurately modelling personalised 3D human digital twins in these simulations, to glean more insights on human-robot interactions. In this paper, we propose to integrate personalised soft-body feet, generated using the motion capture data of real human subjects, into a skeletal model and train it with a walking control policy. Through evaluation using ground reaction force and joint angle results, the soft-body feet were able to generate ground reaction force results comparable to real measured data and closely follow joint angle results of the bare skeletal model and the reference motion. This presents an interesting avenue to produce a dynamically accurate human model in simulation driven by their own control policy while only seeing kinematic information during training.

Paper Structure

This paper contains 13 sections, 1 equation, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Our Proposed Approach
  3. Generating the Feet Model
  4. Integrating Soft-body Feet with the Skeletal Model
  5. Configuring the Material Properties of the Soft-body Feet
  6. Material Property Values
  7. Use of Elasticity Plugin
  8. Results and Analysis
  9. Walking Control Policy Training for the Skeletal Model with Soft-body Feet
  10. Ground Reaction Force (GRF)
  11. Joint Angles
  12. Conclusion and Future Work

Figures (4)

  • Figure 1: Simulation Pipeline
  • Figure 2: Top row: Full body flex model with reduced resolution (Left) and Pin connections between the flex feet and the SK model, as shown by the spheres (Right). Bottom row: From left to right, Original high-resolution foot shape, Foot shape of reduced resolution, Mesh model of the foot, flex object in MuJoCo of the Foot
  • Figure 3: Plot of vertical GRF against gait percentage for models SK, A, B, and E (Top row left to right, bottom row left to right). Note that for models A, B, and E, offsets were applied to better match the GRF profiles.
  • Figure 4: Joint angles for models SK, A, B, and E