Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cycloidal Quasi-Direct Drive Actuator Designs with Learning-based Torque Estimation for Legged Robotics

Alvin Zhu, Yusuke Tanaka, Fadi Rafeedi, Dennis Hong

TL;DR

This work couples a Cycloidal Gear Quasi-Direct Drive (C-QDD) actuator with a learning-based torque estimator to enable high-torque, robust, lightweight actuation for legged robotics while mitigating cycloidal nonlinearities. The design leverages a 10:1 cycloidal gearbox and a GRU-based Actuator Network (with two input variants) to predict nonlinear torque ripple and dynamic behavior, addressing sim-to-real gaps in reinforcement learning contexts. Hardware verified metrics show strong torque density, low backlash, and reasonable backdrivability, with the estimator achieving superior RMSE and ripple tracking compared to MLP baselines. The approach promises improved performance in agile locomotion and climbing tasks, enabling more reliable policy transfer from simulation to real hardware and enabling torque-aware planning and control in challenging terrains.

Abstract

This paper presents a novel approach through the design and implementation of Cycloidal Quasi-Direct Drive actuators for legged robotics. The cycloidal gear mechanism, with its inherent high torque density and mechanical robustness, offers significant advantages over conventional designs. By integrating cycloidal gears into the Quasi-Direct Drive framework, we aim to enhance the performance of legged robots, particularly in tasks demanding high torque and dynamic loads, while still keeping them lightweight. Additionally, we develop a torque estimation framework for the actuator using an Actuator Network, which effectively reduces the sim-to-real gap introduced by the cycloidal drive's complex dynamics. This integration is crucial for capturing the complex dynamics of a cycloidal drive, which contributes to improved learning efficiency, agility, and adaptability for reinforcement learning.

Cycloidal Quasi-Direct Drive Actuator Designs with Learning-based Torque Estimation for Legged Robotics

TL;DR

This work couples a Cycloidal Gear Quasi-Direct Drive (C-QDD) actuator with a learning-based torque estimator to enable high-torque, robust, lightweight actuation for legged robotics while mitigating cycloidal nonlinearities. The design leverages a 10:1 cycloidal gearbox and a GRU-based Actuator Network (with two input variants) to predict nonlinear torque ripple and dynamic behavior, addressing sim-to-real gaps in reinforcement learning contexts. Hardware verified metrics show strong torque density, low backlash, and reasonable backdrivability, with the estimator achieving superior RMSE and ripple tracking compared to MLP baselines. The approach promises improved performance in agile locomotion and climbing tasks, enabling more reliable policy transfer from simulation to real hardware and enabling torque-aware planning and control in challenging terrains.

Abstract

This paper presents a novel approach through the design and implementation of Cycloidal Quasi-Direct Drive actuators for legged robotics. The cycloidal gear mechanism, with its inherent high torque density and mechanical robustness, offers significant advantages over conventional designs. By integrating cycloidal gears into the Quasi-Direct Drive framework, we aim to enhance the performance of legged robots, particularly in tasks demanding high torque and dynamic loads, while still keeping them lightweight. Additionally, we develop a torque estimation framework for the actuator using an Actuator Network, which effectively reduces the sim-to-real gap introduced by the cycloidal drive's complex dynamics. This integration is crucial for capturing the complex dynamics of a cycloidal drive, which contributes to improved learning efficiency, agility, and adaptability for reinforcement learning.

Paper Structure

This paper contains 21 sections, 2 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Actuation Mechanism in Legged Robotics
  4. Cycloidal Gear Mechanisms
  5. Actuator Modeling
  6. Cycloidal Gear Quasi-Direct Drive Mechanisms
  7. Sizing of Cycloidal and Planetary Gears
  8. Cycloid profiles and eccentricity
  9. Non-Pinwheel Cycloid Ring
  10. C-QDD Dynamics and Torque Estimation
  11. Torque Estimation Framework
  12. Model Parameters and Training
  13. Hardware Verification and Analysis
  14. C-QDD Mechanical and Electrical Performance
  15. C-QDD Backlash and Backdrivability Performance
  16. ...and 6 more sections

Figures (7)

  • Figure 1: Proposed cycloidal gear quasi-direct drive actuator and the torque estimation framework.
  • Figure 2: C-QDD actuator design and cycloidal profiles.
  • Figure 3: A cross-section view of the C-QDD design.
  • Figure 4: Overview of the torque estimation framework architecture, leveraging temporal actuator state information to generate precise torque predictions.
  • Figure 5: C-QDD actuator performance benchmark testbed.
  • ...and 2 more figures