A simulation framework for simultaneous design and control of passivity based walkers

Guilhem Saurel, Justin Carpentier, Nicolas Mansard, Jean‐Paul Laumond

Abstract

This paper proposes a simulation framework that simultaneously computes both the physical design and the control of bipedal walkers. Formulated as a single large-scale parametric optimal control problem on hybrid dynamics, the framework leverages state-of-the-art numerical techniques and efficient rigid-body dynamics. The approach allows for the rapid computation of parameterized models and passive walker controls across various scenarios in just a few seconds on standard hardware.

Introduction

Designing passivity-based walkers—robots that exploit natural dynamics and gravity for locomotion—presents a significant "chicken and egg" challenge. A robot's physical design (limb lengths, mass distribution) dictates its natural gait, while its controller must be finely tuned to these specific physics. Traditionally, these two aspects are optimized sequentially, often leading to suboptimal performance or lengthy design cycles. Furthermore, many existing tools are restricted to simple 2D models or are computationally prohibitive for rapid iteration.

This research introduces a high-performance framework that bridges this gap by treating robot design and control as a unified optimization problem. By optimizing both physical parameters and control gains in a single step, the framework can generate energy-efficient walking gaits for complex architectures, including 3D models, with remarkable speed.

The Simulation Framework

The core of the proposed framework is the formulation of the design and control task as a single large-scale parametric optimal control problem. This approach ensures that the physical body and the movement strategy are co-evolved for maximum efficiency.

The simulation framework architecture, showing the interaction between the MUSCOD-II solver and the Pinocchio dynamics engine.

As illustrated in the framework architecture, the system integrates two primary components:

  1. Dynamics Engine (Pinocchio): This C++ library provides high-performance forward and inverse dynamics. It allows the system to automatically derive equations of motion for any kinematic tree, facilitating the exploration of diverse robot architectures without manual derivation.
  2. Optimization Solver (MUSCOD-II): The framework utilizes a Multiple Shooting method. This technique is particularly robust for the hybrid dynamics of walking, which involve continuous swing phases and discrete, inelastic foot-strike impacts. By breaking trajectories into smaller intervals, multiple shooting maintains stability and handles the instabilities inherent in bipedal locomotion better than single shooting or collocation methods.

The framework supports both active control (powered joints using cubic polynomials) and passive control (joints utilizing spring-damper systems).

Experimental Evaluation

To validate the framework, the authors evaluated six distinct walker models ranging from a basic compass walker to a complex 3D model with actuated arms. To maintain biological realism, the framework incorporates anthropometric data to scale mass, inertia, and centers of mass based on reference human dimensions. This ensures that the optimized designs remain physically plausible and "human-like."

Results and Discussion

The framework demonstrated exceptional computational efficiency, with convergence times ranging from 1.2 to 7.6 seconds on standard computing hardware.

Visual output of the framework showing walking sequences and performance metrics (CoT, step length, time) for different robot configurations.

Key findings from the optimization include:

  • Complexity and Efficiency: Increasing the complexity of the robot's morphology often led to significant energy savings. For instance, adding knees to the base compass model reduced the Cost of Transport (CoT) from 0.1007 to 0.0544. The addition of a torso further reduced the CoT by approximately 40%.
  • Design-Control Coupling: The solver successfully identified "sweet spots" where physical parameters (like limb lengths) naturally assisted the gait, thereby reducing the effort required from the actuators.
  • 3D Locomotion: The framework successfully optimized a 3D walker with arms, proving its capability to handle the lateral stability requirements that 2D models typically ignore.

Conclusion

The proposed framework offers a powerful tool for the rapid prototyping of bipedal robots. By unifying design and control optimization, it reveals how morphological complexity can be leveraged to improve energy efficiency.

Future work will focus on incorporating more realistic contact models (e.g., curved feet) and explicitly integrating stability analysis, such as Poincaré maps, to ensure that the generated gaits are robust to external perturbations. While the current implementation uses the closed-source MUSCOD-II solver, the methodology is compatible with open-source alternatives like the ACADO toolkit.

Summary updated

Abstract

In this paper, we propose a simulation framework which simultaneously computes both the design and the control of bipedal walkers. The problem of computing a design and a control is formulated as a single large-scale parametric optimal control problem on hybrid dynamics with path constraints (e.g. non sliding and non slipping contact constraints). Our framework relies on state-of-the-art numerical optimal control techniques and efficient computation of the multi-body rigid dynamics. It allows to compute both the parametrized model and the control of passive walkers on different scenarios, in only few seconds on a standard computer. The framework is illustrated by several examples which highlight the interest of the approach.
BibTeX
@inproceedings{saurel2016simulation,
  author = {Guilhem Saurel and Justin Carpentier and Nicolas Mansard and Jean‐Paul Laumond},
  title = {{A simulation framework for simultaneous design and control of passivity based walkers}},
  booktitle = {SIMPAR},
  year = {2016},
  doi = {10.1109/simpar.2016.7862383},
}
Figure