Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-Driven Dynamic Parameter Learning of manipulator robots

Mohammed Elseiagy, Tsige Tadesse Alemayoh, Ranulfo Bezerra, Shotaro Kojima, Kazunori Ohno

TL;DR

The paper tackles the challenge of accurate dynamic-parameter estimation for robotic manipulators to bridge the sim-to-real gap. It introduces a Transformer-based model trained on a large, automatically generated dataset of 8,192 URDFs enriched with Jacobian-based kinematic features, created through a gravity-aware PID-enabled simulation pipeline. Key findings show near-perfect accuracy for mass and inertia, moderate-to-high accuracy for Coulomb friction, while viscous friction and distal-center-of-mass estimates remain difficult, with performance strongly influenced by sequence length and model capacity. This approach offers a scalable path toward improved model-based control and safer, more reliable sim-to-real transfer in robotic systems.

Abstract

Bridging the sim-to-real gap remains a fundamental challenge in robotics, as accurate dynamic parameter estimation is essential for reliable model-based control, realistic simulation, and safe deployment of manipulators. Traditional analytical approaches often fall short when faced with complex robot structures and interactions. Data-driven methods offer a promising alternative, yet conventional neural networks such as recurrent models struggle to capture long-range dependencies critical for accurate estimation. In this study, we propose a Transformer-based approach for dynamic parameter estimation, supported by an automated pipeline that generates diverse robot models and enriched trajectory data using Jacobian-derived features. The dataset consists of 8,192 robots with varied inertial and frictional properties. Leveraging attention mechanisms, our model effectively captures both temporal and spatial dependencies. Experimental results highlight the influence of sequence length, sampling rate, and architecture, with the best configuration (sequence length 64, 64 Hz, four layers, 32 heads) achieving a validation R2 of 0.8633. Mass and inertia are estimated with near-perfect accuracy, Coulomb friction with moderate-to-high accuracy, while viscous friction and distal link center-of-mass remain more challenging. These results demonstrate that combining Transformers with automated dataset generation and kinematic enrichment enables scalable, accurate dynamic parameter estimation, contributing to improved sim-to-real transfer in robotic systems

Data-Driven Dynamic Parameter Learning of manipulator robots

TL;DR

The paper tackles the challenge of accurate dynamic-parameter estimation for robotic manipulators to bridge the sim-to-real gap. It introduces a Transformer-based model trained on a large, automatically generated dataset of 8,192 URDFs enriched with Jacobian-based kinematic features, created through a gravity-aware PID-enabled simulation pipeline. Key findings show near-perfect accuracy for mass and inertia, moderate-to-high accuracy for Coulomb friction, while viscous friction and distal-center-of-mass estimates remain difficult, with performance strongly influenced by sequence length and model capacity. This approach offers a scalable path toward improved model-based control and safer, more reliable sim-to-real transfer in robotic systems.

Abstract

Bridging the sim-to-real gap remains a fundamental challenge in robotics, as accurate dynamic parameter estimation is essential for reliable model-based control, realistic simulation, and safe deployment of manipulators. Traditional analytical approaches often fall short when faced with complex robot structures and interactions. Data-driven methods offer a promising alternative, yet conventional neural networks such as recurrent models struggle to capture long-range dependencies critical for accurate estimation. In this study, we propose a Transformer-based approach for dynamic parameter estimation, supported by an automated pipeline that generates diverse robot models and enriched trajectory data using Jacobian-derived features. The dataset consists of 8,192 robots with varied inertial and frictional properties. Leveraging attention mechanisms, our model effectively captures both temporal and spatial dependencies. Experimental results highlight the influence of sequence length, sampling rate, and architecture, with the best configuration (sequence length 64, 64 Hz, four layers, 32 heads) achieving a validation R2 of 0.8633. Mass and inertia are estimated with near-perfect accuracy, Coulomb friction with moderate-to-high accuracy, while viscous friction and distal link center-of-mass remain more challenging. These results demonstrate that combining Transformers with automated dataset generation and kinematic enrichment enables scalable, accurate dynamic parameter estimation, contributing to improved sim-to-real transfer in robotic systems

Paper Structure

This paper contains 25 sections, 6 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Classical and Analytical Identification Methods
  4. Deep Learning for Robot Dynamics
  5. Transformers in Robotics and Time-Series Analysis
  6. Methodology
  7. URDF Generator
  8. Dynamic Parameters
  9. Trajectory Generation
  10. Post-Processing and Dataset Composition
  11. Final Dataset Composition
  12. Transformer-Based Model Architecture
  13. Input Embedding
  14. Results
  15. Effect of Dataset Configuration
  16. ...and 10 more sections

Figures (4)

  • Figure 1: The multi-stage pipeline for robotic data generation, simulation, preprocessing, and deep learning analysis.
  • Figure 2: Examples of robots generated by the URDF script. All robots share the same kinematic configuration (link lengths, joint axes, and offsets) but differ in their inertial and frictional dynamics due to changes in link cross-section, diameter, center of mass, and joint friction coefficients.
  • Figure 3: Illustration of the offset-based sampling scheme. Multiple sequences are generated with temporal offsets and overlapping improving dataset utilization.
  • Figure 4: The architecture of the proposed model, based on the original transformer vaswani2017attentionneed.