Bipedal Stepping Controller Design Considering Model Uncertainty: A Data-Driven Perspective.

Area of Science:

• Robotics
• Control Theory
• Artificial Intelligence

Background:

• Bipedal robot locomotion relies on state-feedback controllers for stable walking.
• Reduced-order models (ROMs) are commonly used, but often ignore model discrepancies with full-order systems.
• Addressing model uncertainties is crucial for robust bipedal robot control.

Purpose of the Study:

• To introduce a novel robust stepping controller for bipedal robots.
• To address and overcome model discrepancies ignored in traditional ROM-based controllers.
• To enhance walking robustness against uncertainties and disturbances.

Main Methods:

• Utilized behavioral systems theory to construct a controller directly from input-state data.
• Represented model uncertainties as bounded noise and over-approximated them with bounded energy ellipsoids.
• Employed simulation experiments on a 22-degrees-of-freedom humanoid robot.

Main Results:

• The novel controller demonstrated superior robustness compared to a nominal step-to-step (S2S) controller.
• Successfully handled uncertain loads, various sloped terrains, and push recovery scenarios.
• Validated the effectiveness of using behavioral systems theory for robust robot control.

Conclusions:

• The proposed controller offers a robust solution for bipedal robot locomotion, outperforming traditional methods.
• Directly incorporating real-world data and accounting for model uncertainties leads to enhanced stability and adaptability.
• This approach paves the way for more reliable and versatile humanoid robot navigation.