From Walking to Running: 3D Humanoid Gait Generation via MPC

Area of Science:

• Robotics
• Control Systems
• Humanoid Locomotion

Background:

• Generating stable and dynamic gaits for humanoid robots in real-time remains a significant challenge.
• Existing methods often struggle with adaptability to varying terrains and dynamic maneuvers like running.
• Model Predictive Control (MPC) offers a promising framework for addressing these complexities.

Purpose of the Study:

• To develop a real-time algorithm for stable 3D walking and running in humanoid robots.
• To generate adaptable gaits that closely follow a predefined footstep plan, including Center of Mass (CoM) height and locomotion mode (walking/running).
• To ensure dynamic stability by maintaining the Zero Moment Point (ZMP) within the support region.

Main Methods:

• Utilized a Variable-Height Inverted Pendulum (VH-IP) model for prediction, generating CoM trajectories and adapted footsteps in real-time.
• Handled VH-IP nonlinearity by splitting gait generation into two stages, each solved via quadratic programming.
• Incorporated constraints for Ground Reaction Force (GRF), ZMP stability, kinematic realizability, and bounded CoM evolution relative to ZMP.

Main Results:

• Successfully generated real-time, stable 3D walking and running gaits for humanoid robots.
• Demonstrated the algorithm's ability to adapt footstep plans and CoM trajectories based on the VH-IP model.
• Validated the approach through dynamic simulations on the HRP-4 robot and experiments on the OP3 robot, including running on tilted surfaces.

Conclusions:

• The proposed MPC-based algorithm effectively achieves real-time, stable humanoid locomotion (walking and running) with adaptable gaits.
• The VH-IP model and two-stage quadratic programming approach successfully manage complex dynamics and constraints.
• The method shows potential for robust humanoid robot operation in diverse and challenging environments.