Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rapid Co-design of Task-Specialized Whegged Robots for Ad-Hoc Needs

Varun Madabushi, Katie M. Popek, Craig Knuth, Galen Mullins, Brian A. Bittner

TL;DR

This work tackles rapid, in-field customization of legged robots by co-designing morphology and gait for task-specific terrains. It defines a compact parameter space for six whegs on a MiniRHex and optimizes both morphology and open-loop gait using Bayesian optimization, validated across four terrains with sim-to-real transfer. Hardware experiments show that terrain-optimized designs outperform the nominal platform in efficiency by up to ~5x and generalize from simulation, though speed-optimized gaits may struggle with impulsive contact dynamics. The study supports the viability of in-situ co-design for ad-hoc needs and lays groundwork for fieldable co-design pipelines that quickly yield task-specialist locomotion solutions.

Abstract

In this work, we investigate the use of co-design methods to iterate upon robot designs in the field, performing time sensitive, ad-hoc tasks. Our method optimizes the morphology and wheg trajectory for a MiniRHex robot, producing 3D printable structures and leg trajectory parameters. Tested in four terrains, we show that robots optimized in simulation exhibit strong sim-to-real transfer and are nearly twice as efficient as the nominal platform when tested in hardware.

Rapid Co-design of Task-Specialized Whegged Robots for Ad-Hoc Needs

TL;DR

This work tackles rapid, in-field customization of legged robots by co-designing morphology and gait for task-specific terrains. It defines a compact parameter space for six whegs on a MiniRHex and optimizes both morphology and open-loop gait using Bayesian optimization, validated across four terrains with sim-to-real transfer. Hardware experiments show that terrain-optimized designs outperform the nominal platform in efficiency by up to ~5x and generalize from simulation, though speed-optimized gaits may struggle with impulsive contact dynamics. The study supports the viability of in-situ co-design for ad-hoc needs and lays groundwork for fieldable co-design pipelines that quickly yield task-specialist locomotion solutions.

Abstract

In this work, we investigate the use of co-design methods to iterate upon robot designs in the field, performing time sensitive, ad-hoc tasks. Our method optimizes the morphology and wheg trajectory for a MiniRHex robot, producing 3D printable structures and leg trajectory parameters. Tested in four terrains, we show that robots optimized in simulation exhibit strong sim-to-real transfer and are nearly twice as efficient as the nominal platform when tested in hardware.
Paper Structure (10 sections, 2 equations, 2 figures, 1 table)

This paper contains 10 sections, 2 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Statement
  4. Method
  5. Results
  6. Simulation
  7. Hardware
  8. Conclusions
  9. Supporting Simulation Fidelity of Wheg Compliance
  10. Cross Validation of Optimized co-designs

Figures (2)

  • Figure 1: Four RHex co-designs were optimized in simulation over 100 trials in flat, rough, stairs, and ramp terrains through our approach, described in Section \ref{['sec:method']}. Each design outperforms the nominal gait in efficiency by 1.6x, 2x, 2.2x, and 5.1x respectively on hardware.
  • Figure 2: Efficiencies of each optimized platform on each terrain. All designs equalled or outperformed the nominal design, demonstrating co-design's capability to discover designs that would not be obvious to a human engineer. The optimization process resulted in a terrain-specialized platform which outperformed the nominal and other terrain platforms on its own environment.