Dynamic Walking on Highly Underactuated Point Foot Humanoids: Closing the Loop between HZD and HLIP

Adrian B. Ghansah, Jeeseop Kim, Kejun Li, Aaron D. Ames

TL;DR

This paper presents a novel control framework for achieving dynamic walking on highly underactuated humanoid robots with point feet. By closing the loop between offline gait generation using Hybrid Zero Dynamics (HZD) and online stabilization with a Hybrid Linear Inverted Pendulum (HLIP) regulator, the authors demonstrate robust locomotion, including tracking desired speeds, recovering from pushes, and navigating uneven terrain, marking the first instance of point-foot humanoid walking.

What Problem Does This Paper Address?

Bipedal locomotion is a fundamental challenge for humanoid robots, especially when dealing with highly underactuated systems and point feet. Point feet, lacking passive stabilization like traditional feet, introduce significant complexity due to their hybrid dynamics (continuous motion punctuated by discrete impact events) and high degrees of freedom. Existing methods like Hybrid Zero Dynamics (HZD) can generate stable gaits but are computationally expensive, preventing real-time adjustments. Reduced-order models (ROMs) like the Hybrid Linear Inverted Pendulum (HLIP) offer computational efficiency for online control but often struggle to accurately capture the full-order dynamics of underactuated robots. This paper tackles the gap between these two approaches, aiming to leverage the robustness of HZD for gait generation while utilizing the speed of HLIP for online stabilization, specifically for the challenging case of point-foot humanoids.

Methods

The core of the proposed method is a control framework that integrates two key techniques:

  • Hybrid Zero Dynamics (HZD) Gait Generation: This approach uses trajectory optimization to generate stable, periodic walking gaits offline. A crucial innovation here is the incorporation of the HLIP model's dynamics as constraints during the HZD gait generation process. This ensures that the generated full-order gaits are more aligned with the simplified HLIP model, facilitating smoother online stabilization.
  • Hybrid Linear Inverted Pendulum (HLIP) Regulator: This reduced-order model is used for online stabilization. The HLIP regulator takes the nominal HZD gait trajectories and modifies the swing foot trajectories in real-time to ensure the robot's center of mass (COM) states remain stable. Specifically, it uses two HLIP regulators (one for sagittal, one for frontal plane) to control the swing foot's x and y positions by adjusting step lengths. This allows the robot to adapt to deviations from the nominal gait and maintain stability.

The framework essentially closes the loop by using HZD to generate gaits that are "HLIP-aware," and then using HLIP to actively stabilize these gaits online by adjusting swing foot placement.

Datasets

Not discussed in the paper. The paper focuses on a novel control framework and its experimental validation on a specific robot platform.

Key Insights and Ideas

  • Bridging the Gap: The paper successfully bridges the gap between computationally expensive full-order models (HZD) and computationally efficient reduced-order models (HLIP) for dynamic walking. By making HZD "HLIP-aware" during generation, the subsequent HLIP stabilization becomes more effective.
  • Point Foot Walking is Achievable: Demonstrating dynamic walking on a point-foot humanoid robot is a significant achievement, pushing the boundaries of bipedal locomotion research. This opens doors for understanding control strategies applicable to even more challenging robotic platforms.
  • Robustness through Integration: The integration of HZD and HLIP leads to remarkable robustness. The ability to track desired speeds, recover from pushes, and navigate uneven terrain highlights the effectiveness of this combined approach.
  • Real-time Adaptation: While HZD provides a stable baseline, the HLIP regulator allows for real-time adjustments to step length, enabling the robot to react to disturbances and maintain stability without needing to recompute entire gaits.

Key Results

The paper reports successful validation of the proposed framework through both simulations and hardware experiments on the ADAM humanoid robot.

  • Nominal Walking: The robot was able to walk at various speeds (up to 0.5 m/s forward) and track desired velocity commands, with phase portraits showing stable, periodic COM orbits that closely matched the nominal HZD gaits.
  • Push Recovery: The robot demonstrated robustness by stabilizing itself within four steps after being perturbed by a significant lateral push, returning to its target area.
  • Rough Terrain: The robot successfully traversed uneven terrain (a pile of rubber mats up to 0.11 m high), maintaining velocity tracking capabilities, albeit with some degradation compared to flat ground.
  • First Point Foot Walking: The paper claims to present the first example of dynamic point-foot humanoid walking.

Limitations and Future Work

  • Yaw Slippage: The authors acknowledge challenges with yaw slippage of the point feet due to their rigid metal construction and minimal yaw friction. They suggest that introducing feet with multiple contact points or utilizing actuated ankles could mitigate this.
  • Actuated Ankles: The paper mentions exploring the advantages of actuated ankles as a future direction to further improve locomotion performance.
  • Yaw Control: While the paper focuses on COM and swing foot control, the discussion on yaw slippage implies that more sophisticated yaw control strategies might be beneficial.
  • Model Discrepancies: Although the framework aims to handle model discrepancies, the inherent limitations of ROMs and the complexity of real-world dynamics mean that perfect tracking is not always achieved, as seen in the rough terrain experiments.

Summary updated

Abstract

Realizing bipedal locomotion on humanoid robots with point feet is especially challenging due to their highly underactuated nature, high degrees of freedom, and hybrid dynamics resulting from impacts. With the goal of addressing this challenging problem, this paper develops a control framework for realizing dynamic locomotion and implements it on a novel point foot humanoid: ADAM. To this end, we close the loop between Hybrid Zero Dynamics (HZD) and Hybrid linear inverted pendulum (HLIP) based step length regulation. To leverage the full-order hybrid dynamics of the robot, walking gaits are first generated offline by utilizing HZD. These trajectories are stabilized online through the use of a HLIP based regulator. Finally, the planned trajectories are mapped into the full-order system using a task space controller incorporating inverse kinematics. The proposed method is verified through numerical simulations and hardware experiments on the humanoid robot ADAM marking the first humanoid point foot walking. Moreover, we experimentally demonstrate the robustness of the realized walking via the ability to track a desired reference speed, robustness to pushes, and locomotion on uneven terrain.
BibTeX
@inproceedings{ghansah2024dynamic,
  author = {Adrian B. Ghansah and Jeeseop Kim and Kejun Li and Aaron D. Ames},
  title = {{Dynamic Walking on Highly Underactuated Point Foot Humanoids: Closing the Loop between HZD and HLIP}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2024},
  doi = {10.1109/iros58592.2024.10802542},
}
Figure