Synthesizing Robust Walking Gaits via Discrete-Time Barrier Functions with Application to Multi-Contact Exoskeleton Locomotion

Maegan Tucker, Kejun Li, Aaron D. Ames

TL;DR

This paper introduces a novel method for synthesizing robust walking gaits for bipedal robots by quantifying robustness as the size of a forward-invariant set in the robot's discrete-time dynamics. They demonstrate this approach using simulation-in-the-loop learning and experimentally validate it on a multi-contact exoskeleton, showing improved robustness compared to stability-optimized gaits.

What Problem Does This Paper Address?

Achieving stable and robust bipedal locomotion in robots is a significant challenge. Real-world scenarios involve uncertainties like model inaccuracies, external disturbances, and imperfect state estimation, which can easily lead to falls. While existing methods can generate stable gaits, they often lack formal guarantees of robustness or are computationally expensive to verify. This paper addresses the need for a systematic way to synthesize gaits that are inherently robust to these disturbances, going beyond just achieving stability. They aim to provide a quantifiable measure of robustness that can be directly used in gait optimization.

Methods

The core of this paper's contribution lies in framing locomotive robustness through the lens of forward invariance of the robot's discrete-time step-to-step dynamics.

  • Discrete-Time Dynamics: They first model the robot's hybrid dynamics (continuous motion during swing, discrete impact at foot contact) as a discrete-time system using the Poincaré return map. This simplifies the analysis of periodic walking.
  • Forward Invariance as Robustness Metric: The key insight is that a larger forward-invariant set around the nominal walking gait's fixed point corresponds to a larger region of attraction and thus greater robustness.
  • Discrete-Time Barrier Functions: To formally certify forward invariance and estimate the size of these sets, they employ discrete-time barrier functions. These functions provide a condition that, if met, guarantees that the system will remain within a safe set.
  • Simulation-in-the-Loop Learning: They develop a simulation-in-the-loop learning framework to optimize gaits. This involves:
  • Generating candidate gaits.
  • Simulating them to find the fixed point of the discrete-time dynamics.
  • Using perturbation analysis and the barrier function method to estimate the largest forward-invariant set (quantified by its radius, $r^*$).
  • Using a preference-based learning algorithm to select gaits that maximize $r^*$.
  • Reduced-Order Representation: For computational tractability, they often work with a reduced-order representation of the system's state, focusing on key variables like global angular velocities, which are believed to be highly influential on stability and robustness.

Datasets

Not discussed in the paper. The experiments are conducted using simulations (MuJoCo) and on a physical robot (Atalante exoskeleton).

Key Insights and Ideas

  • Robustness is more than stability: The paper convincingly argues that optimizing solely for stability can lead to gaits that fail on hardware, highlighting the critical need for explicit robustness metrics.
  • Forward invariance as a quantifiable robustness measure: By relating the size of the forward-invariant set to robustness, they provide a principled and computable metric that can guide gait synthesis.
  • Discrete-time barrier functions are powerful tools for safety and robustness: They demonstrate the practical utility of these functions for certifying forward invariance in complex hybrid systems, even in high-dimensional settings.
  • Simulation-in-the-loop learning is effective for complex optimization: Their framework efficiently explores the gait parameter space to find gaits that are both stable and robust.
  • Multi-contact locomotion benefits significantly from this approach: The successful application to multi-contact walking, which involves more complex domain transitions, underscores the generality of their method.

Key Results

  • The proposed method successfully synthesized robust nominal limit cycles for both flat-foot and multi-contact walking on the Atalante exoskeleton.
  • Experimental results showed that gaits optimized for stability alone were not robust enough for hardware deployment, whereas gaits optimized using their robustness metric ($r^*$) were successful.
  • They identified invariant sets and evaluated barrier functions using experimental data, visually demonstrating the robustness of the synthesized gaits.
  • The framework is computationally tractable, allowing for the optimization of gaits in high-dimensional systems.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that robustness is not the only factor for desirable locomotion. Other aspects like user comfort and naturalness are also important.
  • The current work focuses on nominal gaits; integrating this robustness metric with online planners could further enhance performance.
  • Future Work:
    • Combine robustness with other important metrics like user comfort and naturalness.
  • Explore the application of this framework to reinforcement learning settings as part of reward composition.
  • Investigate the impact of different reduced-order representations on the effectiveness of the robustness metric.

Summary updated

Abstract

Successfully achieving bipedal locomotion remains challenging due to real-world factors such as model uncertainty, random disturbances, and imperfect state estimation. In this work, we propose a novel metric for locomotive robustness -- the estimated size of the hybrid forward invariant set associated with the step-to-step dynamics. Here, the forward invariant set can be loosely interpreted as the region of attraction for the discrete-time dynamics. We illustrate the use of this metric towards synthesizing nominal walking gaits using a simulation-in-the-loop learning approach. Further, we leverage discrete-time barrier functions and a sampling-based approach to approximate sets that are maximally forward invariant. Lastly, we experimentally demonstrate that this approach results in successful locomotion for both flat-foot walking and multi-contact walking on the Atalante lower-body exoskeleton.
BibTeX
@inproceedings{tucker2024synthesizing,
  author = {Maegan Tucker and Kejun Li and Aaron D. Ames},
  title = {{Synthesizing Robust Walking Gaits via Discrete-Time Barrier Functions with Application to Multi-Contact Exoskeleton Locomotion}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2024},
  doi = {10.1109/icra57147.2024.10610537},
}
Figure