Biped Walking Based on Stiffness Optimization and Hierarchical Quadratic Programming

Area of Science:

• Robotics
• Control Systems
• Biomechanics

Background:

• The spring-loaded inverted pendulum model is a standard for humanoid robot motion planning due to similarities with human walking.
• Existing control methods like velocity feedback and nonlinear optimization have limitations in accuracy, complexity, and real-time application.

Purpose of the Study:

• To develop a real-time optimization method for humanoid robot motion control.
• To improve the stability and disturbance rejection capabilities of biped robots on uneven terrain.

Main Methods:

• Utilized a center of mass (CoM)-velocity feedback controller to determine a virtual landing point.
• Constructed a touchdown return map and employed nonlinear least squares for spring stiffness optimization.
• Implemented hierarchical quadratic programming for whole-body control with prioritized dynamic equations.

Main Results:

• The proposed method enables stable walking on unknown uneven ground with obstacles up to 5 cm.
• The force-controlled biped robot successfully recovered from a 5 Nm disturbance without falling.
• Reduced the number of optimizations by directly using inverse dynamics for the highest priority dynamic equation.

Conclusions:

• The developed CoM-velocity feedback and nonlinear least squares optimization method allows for real-time control of biped robots.
• The hierarchical quadratic programming approach ensures efficient whole-body control with task prioritization.
• The system demonstrates robust performance in challenging walking scenarios, including uneven terrain and external disturbances.