Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DABI: Evaluation of Data Augmentation Methods Using Downsampling in Bilateral Control-Based Imitation Learning with Images

Masato Kobayashi, Thanpimon Buamanee, Yuki Uranishi

TL;DR

The paper addresses data scarcity and asynchronous sensing in bilateral control-based imitation learning for robotic manipulation. It introduces DABI, which collects high-frequency robot data at $1000\,\text{Hz}$ and images at $100\,\text{Hz}$, downsamples to $100\,\text{Hz}$, and augments the dataset using equidistant data around image acquisitions, enabling up to a tenfold increase in data for Bi-ACT. Experiments on a Put-in-Drawer task show DABI achieves 100% success across trained and untrained objects, outperforming simple downsampling and non-DABI augmentation approaches. The results demonstrate a robust, image-informed data augmentation strategy that reduces the required expert demonstrations and improves generalization in bilateral control imitation learning.

Abstract

Autonomous robot manipulation is a complex and continuously evolving robotics field. This paper focuses on data augmentation methods in imitation learning. Imitation learning consists of three stages: data collection from experts, learning model, and execution. However, collecting expert data requires manual effort and is time-consuming. Additionally, as sensors have different data acquisition intervals, preprocessing such as downsampling to match the lowest frequency is necessary. Downsampling enables data augmentation and also contributes to the stabilization of robot operations. In light of this background, this paper proposes the Data Augmentation Method for Bilateral Control-Based Imitation Learning with Images, called "DABI". DABI collects robot joint angles, velocities, and torques at 1000 Hz, and uses images from gripper and environmental cameras captured at 100 Hz as the basis for data augmentation. This enables a tenfold increase in data. In this paper, we collected just 5 expert demonstration datasets. We trained the bilateral control Bi-ACT model with the unaltered dataset and two augmentation methods for comparative experiments and conducted real-world experiments. The results confirmed a significant improvement in success rates, thereby proving the effectiveness of DABI. For additional material, please check https://mertcookimg.github.io/dabi

DABI: Evaluation of Data Augmentation Methods Using Downsampling in Bilateral Control-Based Imitation Learning with Images

TL;DR

The paper addresses data scarcity and asynchronous sensing in bilateral control-based imitation learning for robotic manipulation. It introduces DABI, which collects high-frequency robot data at and images at , downsamples to , and augments the dataset using equidistant data around image acquisitions, enabling up to a tenfold increase in data for Bi-ACT. Experiments on a Put-in-Drawer task show DABI achieves 100% success across trained and untrained objects, outperforming simple downsampling and non-DABI augmentation approaches. The results demonstrate a robust, image-informed data augmentation strategy that reduces the required expert demonstrations and improves generalization in bilateral control imitation learning.

Abstract

Autonomous robot manipulation is a complex and continuously evolving robotics field. This paper focuses on data augmentation methods in imitation learning. Imitation learning consists of three stages: data collection from experts, learning model, and execution. However, collecting expert data requires manual effort and is time-consuming. Additionally, as sensors have different data acquisition intervals, preprocessing such as downsampling to match the lowest frequency is necessary. Downsampling enables data augmentation and also contributes to the stabilization of robot operations. In light of this background, this paper proposes the Data Augmentation Method for Bilateral Control-Based Imitation Learning with Images, called "DABI". DABI collects robot joint angles, velocities, and torques at 1000 Hz, and uses images from gripper and environmental cameras captured at 100 Hz as the basis for data augmentation. This enables a tenfold increase in data. In this paper, we collected just 5 expert demonstration datasets. We trained the bilateral control Bi-ACT model with the unaltered dataset and two augmentation methods for comparative experiments and conducted real-world experiments. The results confirmed a significant improvement in success rates, thereby proving the effectiveness of DABI. For additional material, please check https://mertcookimg.github.io/dabi
Paper Structure (18 sections, 2 equations, 11 figures, 2 tables)

This paper contains 18 sections, 2 equations, 11 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. Related Works
  3. Bilateral Control-Based Imitation Learning
  4. Data Augmentation Method for Robot Learning
  5. Control System
  6. Controller
  7. Bilateral Control
  8. DABI: Data Augmentation Method for Bilateral Control-Based Imitation Learning with Images
  9. Overview
  10. Data Collection
  11. Data Augmentation
  12. Experiments
  13. Hardware
  14. Environment Setting
  15. Learning Architecture
  16. ...and 3 more sections

Figures (11)

  • Figure 1: Overview of Data Augmentation Method for Bilateral Control-Based Imitation Learning with Images (DABI)
  • Figure 2: Block Diagram of Bilateral Control-Based Imitation Learning
  • Figure 3: Block Diagram of Robot Control System
  • Figure 4: DABI: Data Augmentation Method for Bilateral Control-Based Imitation Learning with Images
  • Figure 5: Image Diagram of Data Augmentation in DABI
  • ...and 6 more figures