Bipedal Walking on Constrained Footholds: Momentum Regulation via Vertical COM Control

Min Dai, Xiaobin Xiong, Aaron D. AmesICRA 2022 · DOI

TL;DR

This paper presents a novel online control framework for bipedal robots to achieve stable walking on challenging terrains with constrained footholds (like stairs and stepping stones). By regulating the robot's angular momentum through precise control of the vertical Center of Mass (COM) velocity, the method enables dynamic and robust locomotion without relying on pre-computed gait libraries.

What Problem Does This Paper Address?

Walking on uneven terrain with discrete footholds is a fundamental challenge for bipedal robots. Traditional methods often struggle with underactuated robots (robots with fewer actuators than degrees of freedom) on such terrains, especially when footholds are constrained in size or position. Existing approaches, like those based on Hybrid Zero Dynamics (HZD), provide formal stability guarantees but are sensitive to perturbations and don't generalize well to arbitrary footholds. Other methods using reduced-order models (like the Linear Inverted Pendulum - LIP) are better suited for dynamic walking but don't inherently handle constrained footholds. This paper aims to bridge this gap by developing a unified framework that combines the strengths of both approaches, enabling dynamic, reactive, and stable walking on complex terrains.

Methods

The core of the proposed method lies in a two-pronged approach:

  • Underactuated Dynamics Approximation: The paper uses the Linear Inverted Pendulum (LIP) model as a reduced-order model to approximate the robot's continuous walking dynamics. This simplification allows for easier analysis and control design, particularly for regulating the robot's angular momentum.
  • Online Planning and Control:
    • Output Synthesis: Desired trajectories for key robot outputs (like torso angle, swing foot position, and vertical COM velocity) are generated. For most outputs, closed-form polynomials are used, with coefficients determined by the desired pre-impact posture and the current step's constraints.
  • Vertical COM Control: The crucial innovation is regulating the vertical COM velocity to achieve desired post-impact angular momentum. This is accomplished using an online, shrinking-horizon Model Predictive Control (MPC) approach. The MPC optimizes the vertical COM trajectory to meet specific pre-impact conditions, ensuring stable transitions between steps.
  • Feedback Controller: A task-space Quadratic Program (QP) based controller is employed to track the synthesized trajectories while respecting physical constraints like motor torque limits and ground contact forces.

Datasets

Not discussed in the paper. The paper focuses on simulation and theoretical development, with plans for hardware verification.

Key Insights and Ideas

  • Momentum Regulation via Vertical COM: The key insight is that by precisely controlling the vertical COM velocity before impact, one can effectively regulate the robot's angular momentum after impact. This is a powerful way to stabilize underactuated dynamics during discrete events.
  • Unified Framework for Constrained Footholds: The paper successfully integrates concepts from dynamic walking (LIP model) with the challenges of constrained footholds, offering a more generalizable solution than previous methods.
  • Online, Reactive Planning: The use of MPC for vertical COM trajectory planning allows the robot to react to upcoming footholds in real-time, making it suitable for dynamic and unpredictable environments.
  • Robustness to Model Differences: The approach demonstrates effectiveness on two very different robots (planar AMBER and 3D Cassie), suggesting its generality across various bipedal platforms.

Key Results

The proposed method was successfully demonstrated in simulation on both the AMBER and 3D Cassie robots.

  • Stable Walking: Both robots were able to walk stably on flat ground, stairs (ascending and descending), and randomly located stepping stones with varying distances and heights.
  • Dynamic Range: The system achieved a wide range of COM forward speeds, from 0.4m/s to 1.6m/s on AMBER.
  • Robustness: The approach showed robustness to varying step durations and stone configurations, with successful traversal over 1000 steps on randomly generated stepping stones for AMBER.
  • Real-time Performance: The MPC and QP controllers were shown to be solvable at a 1kHz frequency, crucial for real-time robotic control.

Limitations and Future Work

  • 3D Coronal Plane: While the framework is applied to 3D Cassie, the current implementation primarily focuses on sagittal plane stabilization. The coronal plane control assumes wide footholds. Future work aims to plan coordination between sagittal and coronal planes for true 3D stepping stone locomotion.
  • Hardware Verification: The paper acknowledges that experimental results on hardware are pending, though the planning and control approach has been implemented on Cassie's secondary PC.
  • Parameter Tuning: The stability of walking on stairs was found to be sensitive to the parameter $\epsilon$. Further understanding and a corresponding feedback strategy for choosing $\epsilon$ based on stone configurations are needed.
  • Unified Momentum Regulation: The authors suggest unifying the characterization of walking with both positive and negative orbital energy for more comprehensive 3D momentum regulation.

Summary updated

BibTeX
@inproceedings{dai2022bipedal,
  author = {Min Dai and Xiaobin Xiong and Aaron D. Ames},
  title = {{Bipedal Walking on Constrained Footholds: Momentum Regulation via Vertical COM Control}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2022},
  doi = {10.1109/icra46639.2022.9812247},
}
Figure