Actuator design of compliant walkers via optimal control

Gabriele Buondonno, Justin Carpentier, Guilhem Saurel, Nicolas Mansard, Alessandro De Luca, Jean-Paul LaumondIROS 2017 · DOI

Abstract

This research introduces a co-design optimization framework that simultaneously determines optimal physical hardware parameters and efficient walking trajectories for bipedal robots. By integrating the design of Series Elastic Actuators (SEA) and Parallel Elastic Actuators (PEA) into an optimal control problem, the study demonstrates that hardware-software co-optimization can yield massive improvements in energy efficiency compared to traditional rigid-joint architectures.

Introduction

Designing energy-efficient walking robots remains a fundamental challenge in robotics. While biological systems naturally exploit passive dynamics and muscle elasticity to minimize energy expenditure, most contemporary robots rely on rigid actuators that often work against natural forces.

To bridge this gap, engineers have explored compliant actuators, such as Series Elastic Actuators (SEA) and Parallel Elastic Actuators (PEA). However, determining the optimal mechanical properties (e.g., spring stiffness) and corresponding control strategies is complex and traditionally relies on trial-and-error. This paper addresses the need for a unified mathematical framework that optimizes both the robot's hardware parameters and its motion trajectories concurrently.

Methodology

The proposed co-design framework is built upon Optimal Control Problems (OCP). Unlike traditional methods that optimize paths for fixed hardware, this approach treats physical parameters—such as spring stiffness ($K$) and motor inertia ($J$)—as decision variables.

Technical Implementation

  • Dynamic Modeling: Lagrangian formulations were developed for three actuator configurations: Rigid, SEA, and PEA.
  • Optimization Solver: The framework employs MUSCOD-II, a solver utilizing the multiple shooting method, which is highly effective for complex, constrained dynamic systems.
  • Software Stack: Body dynamics are computed using Pinocchio, a high-performance C++ library for polyarticulated systems.
  • Objective Functions: Two primary metrics were evaluated: the Cost of Transport (CoT) for energy efficiency and the Squared Norm of Torque as a proxy for natural, low-stress movement.

Evaluation Models

The framework was tested on two distinct simulated models:

  1. Planar Compass Walker: A 2-Degree-of-Freedom (DoF) model consisting of two legs and a hip.
  2. Full 3D Humanoid: A 12-DoF model with human-like mass distribution, including knees, hips, and upper body joints.

Key Insights and Discussion

The study reveals that the benefits of elasticity are only fully realized when the walking gait is specifically tailored to the spring's properties. The optimizer identifies strategies where motor torque is minimized by leveraging the spring's potential energy.

SEA vs. PEA Characteristics

  • SEAs enable extreme efficiency but can induce high-frequency oscillations in motor shafts. They typically require higher stiffness values to be effective.
  • PEAs serve to "unload" the motor by carrying a portion of the robot's weight, effectively bringing the average required torque closer to zero.

Furthermore, minimizing the squared norm of torque leads to the emergence of more "human-like" gaits compared to simple work minimization. Notably, the OCP solver can find optimal designs for a complex 3D humanoid in approximately one minute.

Key Results

Quantitative analysis in simulation showed dramatic performance gains through co-design.

Comparison of gait trajectories and performance metrics for rigid, SEA, and parallel elastic actuators across compass and humanoid models.

As shown in the table above, the "Cost" (squared torque) for the compass walker was reduced from 0.09563 in the rigid configuration to just 0.00004 with an SEA—a near-total reduction in motor effort. For the full humanoid, the Cost of Transport (CoT) dropped from 0.51831 (rigid) to 0.00844 (SEA). The framework successfully identified optimal stiffness values, such as ~1254 Nm/rad for the compass walker's hip, which would be difficult to determine through manual tuning.

Limitations and Future Work

While the framework is powerful, several areas for improvement remain:

  • Automated Actuator Selection: Future iterations could use Mixed-Integer Optimal Control (MIOCP) to automatically decide between SEA, PEA, or rigid joints for each specific articulation.
  • Model Fidelity: Incorporating low-level electrical constraints (e.g., current limits) and accounting for the mass penalties of larger springs would improve real-world applicability.
  • Dynamic Range: Expanding the scope beyond steady-state walking to include running and jumping would further demonstrate the versatility of the co-design approach.

Summary updated

BibTeX
@inproceedings{buondonno2017actuator,
  author = {Gabriele Buondonno and Justin Carpentier and Guilhem Saurel and Nicolas Mansard and Alessandro De Luca and Jean-Paul Laumond},
  title = {{Actuator design of compliant walkers via optimal control}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2017},
  doi = {10.1109/iros.2017.8202228},
}
Figure