Orbit Characterization, Stabilization and Composition on 3D Underactuated Bipedal Walking via Hybrid Passive Linear Inverted Pendulum Model

Xiaobin Xiong, Aaron D. AmesIROS 2019 · DOI

TL;DR

This paper introduces a novel Hybrid-Linear Inverted Pendulum (H-LIP) model to formally characterize and stabilize periodic gaits for 3D underactuated bipedal robots. By synthesizing velocity-based stepping controllers derived from this model, the authors demonstrate a method for generating and smoothly transitioning between various walking gaits, including complex 3D motions, without requiring extensive parameter tuning.

What Problem Does This Paper Address?

Bipedal walking, especially in 3D and for underactuated robots like Cassie, is a complex control problem. Traditional approaches often rely on trajectory optimization, which can be computationally expensive, prone to local minima, and challenging to stabilize online. Simplified models like the Linear Inverted Pendulum (LIP) and Spring-Loaded Inverted Pendulum (SLIP) have been used as approximations or templates, but directly embedding their dynamics into underactuated robots is problematic due to limited control authority. Existing methods for stabilizing gaits using simplified models often rely on heuristics and manual parameter tuning. This paper aims to bridge this gap by providing a formal framework for characterizing, stabilizing, and composing periodic walking gaits for 3D underactuated robots using a novel simplified model.

Methods

The core of this work is the introduction of the Hybrid-Linear Inverted Pendulum (H-LIP) model. This model extends the passive LIP by incorporating two support phases (Single Support Phase - SSP, and Double Support Phase - DSP) and fixed domain durations. The authors formally characterize the periodic orbits of this H-LIP model, specifically Period-1 (P1) and Period-2 (P2) orbits, by analyzing their behavior in the state space.

Based on this characterization, they derive velocity-based stepping controllers designed for global stabilization of these orbits. A key contribution is the derivation of optimal feedback gains for these controllers, which eliminates the need for manual parameter tuning. The paper then demonstrates how these H-LIP stepping controllers can be applied to more complex models:

  1. Actuated Spring-Loaded Inverted Pendulum (aSLIP): The H-LIP stepping is used to stabilize and generate various walking gaits from a stepping-in-place motion.
  2. 3D Underactuated Robot (Cassie): The H-LIP stepping is extended to 3D by composing P1 and P2 orbits in the sagittal and coronal planes, leading to four distinct 3D walking gait compositions. These compositions are then stabilized and generated from a stepping-in-place motion.

Datasets

Not discussed in the paper. The paper focuses on simulation and theoretical analysis using established robot models (aSLIP, Cassie) and their dynamics.

Key Insights and Ideas

  • H-LIP as a Unified Framework: The H-LIP model provides a unified framework for characterizing and stabilizing periodic gaits, acting as a powerful abstraction for complex bipedal locomotion.
  • Formal Orbit Characterization: The paper formally characterizes P1 and P2 orbits of the H-LIP, providing analytical insights into their behavior and stability regions.
  • Optimal, Tunable-Free Stabilization: The derived stepping controllers with optimal gains offer a significant advantage by eliminating the need for heuristic tuning, which is common in many bipedal walking control methods.
  • Compositionality for 3D Walking: The ability to compose P1 and P2 orbits in different planes (sagittal and coronal) offers a systematic way to generate and categorize a wide range of 3D walking gaits.
  • Smooth Transition from Stepping-in-Place: The method effectively stabilizes and generates walking gaits from a simple stepping-in-place motion, demonstrating its practical utility.

Key Results

  • The H-LIP model's P1 and P2 orbits are geometrically characterized, and stepping controllers are designed for their global stabilization.
  • Valid ranges for feedback gains are derived, and optimal gains are identified.
  • The H-LIP stepping successfully stabilizes and generates various P1 and P2 orbits for the aSLIP model from a stepping-in-place motion, achieving low velocity errors.
  • For the 3D Cassie robot, the H-LIP stepping enables the generation and stabilization of four distinct 3D walking gait compositions (sP1-cP1, sP1-cP2, sP2-cP1, sP2-cP2) from a stepping-in-place motion to different target velocities.
  • Simulation results show that the converged orbits for Cassie closely match the predicted H-LIP orbits, and velocities and step sizes converge quickly to desired values.

Limitations and Future Work

  • Theoretical Guarantee for Complex Models: The authors acknowledge that a theoretical guarantee for stabilization from the H-LIP to complex hybrid models of walking is yet to be established. They posit that the hybrid nature of alternating support legs is key.
  • Kinematic Constraints: The paper notes that avoiding kinematic conflicts (e.g., step location relative to stance, legs crossing) is out of the scope of this work, though it's a crucial aspect for real-world robots.
  • Extension to Other Robots: The authors express interest in extending the H-LIP stepping to other types of walking robots, including those with different dynamics or locomotion styles.
  • Further Exploration of Gait Categorization: The proposed orbit composition offers a metric for categorizing bipedal gaits, which could be further explored.
  • Continuous Gait Synthesis: The approach of connecting and continuously stabilizing orbits from a single stepping-in-place optimization is highlighted as a promising direction for continuous gait synthesis.

Summary updated

BibTeX
@inproceedings{xiong2019orbit,
  author = {Xiaobin Xiong and Aaron D. Ames},
  title = {{Orbit Characterization, Stabilization and Composition on 3D Underactuated Bipedal Walking via Hybrid Passive Linear Inverted Pendulum Model}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2019},
  doi = {10.1109/iros40897.2019.8968162},
}
Figure