First Steps Towards Full Model Based Motion Planning and Control of Quadrupeds: A Hybrid Zero Dynamics Approach

Wen-Loong Ma, Kaveh Akbari Hamed, Aaron D. AmesIROS 2019 · DOI

TL;DR

This paper extends the Hybrid Zero Dynamics (HZD) framework, previously successful for bipedal robots, to quadrupedal locomotion. The authors present a unified approach for planning and controlling gaits like walking, ambling, and trotting on a full-model quadruped, achieving stable locomotion experimentally. This work bridges the gap between theoretical HZD and practical quadrupedal robotics, paving the way for more dynamic and robust behaviors.

What Problem Does This Paper Address?

Quadrupedal robots are known for their agility and dynamic stability, but achieving this through formal control and planning methods has been challenging. Existing approaches often rely on simplified models (e.g., massless legs, linear inverted pendulums), which can lead to controllers that struggle with real-world uncertainties and lack formal stability guarantees. The inherent complexity of quadrupedal dynamics, with more degrees of freedom and richer contact scenarios compared to bipeds, has made it difficult to apply advanced control techniques like Hybrid Zero Dynamics (HZD). This paper aims to overcome these challenges by developing a comprehensive framework for full-model based motion planning and control of quadrupeds, leveraging the power of HZD to provide provable stability.

Methods

The authors propose a framework that systematically generates and stabilizes optimal quadrupedal trajectories. The core methodology involves:

  • Hybrid Dynamical Modeling: The quadruped's locomotion is modeled as a hybrid system, capturing the alternating continuous-time dynamics (during stance and swing phases) and discrete-time dynamics (during foot impacts and lift-offs). This allows for a more accurate representation of the robot's behavior.
  • Gait Planning via Nonlinear Programming (NLP): A scalable nonlinear programming approach, using direct collocation and the HZD framework, is employed to generate a wide range of optimal trajectories for gaits like walking, ambling, and trotting. The optimization aims to minimize control effort while satisfying physical constraints.
  • Exponential Stabilization with BMI: To guarantee exponential stability of the generated gaits, an iterative algorithm is used. This algorithm leverages Poincaré sections analysis and designs controllers using a combination of Linear Matrix Inequalities (LMIs) and Bilinear Matrix Inequalities (BMIs). This BMI-based approach allows for the synthesis of controllers that ensure the stability of the fixed points of the Poincaré return map.
  • Input-Output Feedback Linearization: For tracking the planned trajectories, an input-output feedback linearization controller is implemented.

Datasets

Not discussed in the paper. The paper focuses on developing a general framework and validating it with simulations and experiments on a specific robot.

Key Insights and Ideas

  • HZD Extends to Quadrupeds: The paper demonstrates that the Hybrid Zero Dynamics (HZD) framework, which has been highly successful for bipedal robots, can be effectively extended to the more complex domain of quadrupedal locomotion.
  • Unified Framework for Multiple Gaits: The proposed method provides a unified approach to generate and stabilize diverse gaits (walk, amble, trot) by simply adjusting foot clearance constraints within the optimization.
  • BMI for Provable Stability: The use of BMI optimization is crucial for translating the theoretical concept of exponential stability into concrete controller parameters, offering formal guarantees that are often missing in other quadrupedal control methods.
  • Full-Model Approach: By working with the full-order nonlinear and hybrid dynamical models, the approach avoids the limitations of model reduction and aims for controllers that are more robust to real-world dynamics.

Key Results

  • Successful Gait Generation: The framework successfully generates feasible trajectories for walking, ambling, and trotting gaits in simulation.
  • Experimental Validation: The ambling gait was experimentally validated on the Vision 60 quadruped robot, demonstrating stable locomotion at a speed of 0.3 m/s.
  • Stabilization of Trotting Gait: The BMI optimization successfully stabilized a 4-domain trotting gait, transforming dominant eigenvalues from unstable to stable values, and this was visualized through phase portraits.
  • Robustness Demonstrated: The experimental ambling, despite the absence of explicit feedback for uneven terrain, showed remarkable stability, supporting the feasibility of the full-model based HZD approach.

Limitations and Future Work

  • Computational Cost: The BMI optimization, while effective, is computationally intensive, with each iteration taking approximately 30 minutes on their hardware.
  • Drifting in Experiments: The experimental ambling exhibited consistent drifting aside, attributed to the lack of feedback for uneven terrain and potential asymmetric weight distribution.
  • Future Work:
    • Formally defining robustness for quadrupedal locomotion.
  • Designing optimal controllers that can allocate force through actuators for frequently encountered overactuated scenarios.
  • Exploring more dynamic behaviors beyond walking, ambling, and trotting.

Summary updated

BibTeX
@inproceedings{ma2019first,
  author = {Wen-Loong Ma and Kaveh Akbari Hamed and Aaron D. Ames},
  title = {{First Steps Towards Full Model Based Motion Planning and Control of Quadrupeds: A Hybrid Zero Dynamics Approach}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2019},
  doi = {10.1109/iros40897.2019.8968189},
}
Figure