Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Highly Dynamic Behaviors for Quadrupedal Robots

Chong Zhang, Jiapeng Sheng, Tingguang Li, He Zhang, Cheng Zhou, Qingxu Zhu, Rui Zhao, Yizheng Zhang, Lei Han

TL;DR

A learning-based approach to enable robots to learn highly dynamic behaviors from animal motion data is proposed and the results show that the controller is able to reproduce highly dynamic behaviors including sprinting, jumping and sharp turning.

Abstract

Learning highly dynamic behaviors for robots has been a longstanding challenge. Traditional approaches have demonstrated robust locomotion, but the exhibited behaviors lack diversity and agility. They employ approximate models, which lead to compromises in performance. Data-driven approaches have been shown to reproduce agile behaviors of animals, but typically have not been able to learn highly dynamic behaviors. In this paper, we propose a learning-based approach to enable robots to learn highly dynamic behaviors from animal motion data. The learned controller is deployed on a quadrupedal robot and the results show that the controller is able to reproduce highly dynamic behaviors including sprinting, jumping and sharp turning. Various behaviors can be activated through human interaction using a stick with markers attached to it. Based on the motion pattern of the stick, the robot exhibits walking, running, sitting and jumping, much like the way humans interact with a pet.

Learning Highly Dynamic Behaviors for Quadrupedal Robots

TL;DR

A learning-based approach to enable robots to learn highly dynamic behaviors from animal motion data is proposed and the results show that the controller is able to reproduce highly dynamic behaviors including sprinting, jumping and sharp turning.

Abstract

Learning highly dynamic behaviors for robots has been a longstanding challenge. Traditional approaches have demonstrated robust locomotion, but the exhibited behaviors lack diversity and agility. They employ approximate models, which lead to compromises in performance. Data-driven approaches have been shown to reproduce agile behaviors of animals, but typically have not been able to learn highly dynamic behaviors. In this paper, we propose a learning-based approach to enable robots to learn highly dynamic behaviors from animal motion data. The learned controller is deployed on a quadrupedal robot and the results show that the controller is able to reproduce highly dynamic behaviors including sprinting, jumping and sharp turning. Various behaviors can be activated through human interaction using a stick with markers attached to it. Based on the motion pattern of the stick, the robot exhibits walking, running, sitting and jumping, much like the way humans interact with a pet.
Paper Structure (10 sections, 8 equations, 10 figures)

This paper contains 10 sections, 8 equations, 10 figures.

Table of Contents

  1. INTRODUCTION
  2. METHODOLOGY
  3. Data Acquisition and Processing
  4. Motion Imitation
  5. Gait Activation
  6. Real World Transfer
  7. EXPERIMENTAL RESULTS
  8. Motion Imitation
  9. Gait Activation
  10. CONCLUSIONS

Figures (10)

  • Figure 1: The framework of proposed model. It consists of two stages: (a) motion imitation and (b) gait activation. In motion imitation stage, the model is trying to imitate animal behaviors. In gait activation stage, humans can interact with the robot by triggering various gaits. The module in red dashed rectangle in (a) is replaced by module in red solid rectangle in (b), the remaining module is shared by (a) and (b).
  • Figure 2: Statistics of collected motion data from a medium-sized Labrador retriever. Horizontal axis represents gait type, vertical axis represents the duration in minutes.
  • Figure 3: Snapshots of learned robot behaviors. (a) Snapshots of sprinting. (b) Snapshots of jumping
  • Figure 4: The evolution of imitation performance. Horizontal axis represents training time in hours, vertical axis represents average return. Colors represent different gaits.
  • Figure 5: Joint angle curves tested at the end of training. Only curves from front left leg are shown for clarity. Top, mid and bottom panel represent curves of representative walking, running and jumping motion. Horizontal axis represents time in second and vertical axis represent joint angle in radian. Solid lines represents reference data, dashed lines represent the joint angles of real robot. Shaded areas represents the standard deviation computed over three trials. Hip joint, upper leg joint and knee joint are colored with red, blue and green respectively.
  • ...and 5 more figures