Adapting stiffness and attack angle through trial and error to increase self-stability in locomotion

Area of Science:

• Robotics
• Biomechanics
• Control Theory

Background:

• Biological systems exhibit superior legged locomotion compared to machines.
• Adaptability in control and morphology is key to biological locomotion success.
• Existing robotic locomotion models often lack adaptive learning capabilities.

Purpose of the Study:

• To extend the spring-loaded inverted pendulum (SLIP) model with adaptive learning.
• To investigate how adapting attack angle (control) and stiffness (morphology) impacts locomotion stability.
• To enable robotic systems to learn from past locomotion attempts for improved performance.

Main Methods:

• Systematic investigation of various update rules for adaptation.
• Extension of the SLIP model to incorporate adjustable attack angle and stiffness.
• Simulations and analysis of locomotion stability under different adaptation strategies.

Main Results:

• Adapting either attack angle or stiffness, or both, significantly improves locomotion stability.
• Increasing attack angle and decreasing stiffness were found to be beneficial for stability when performance improves.
• The extended SLIP model demonstrated stable locomotion across a wider parameter range via trial and error.

Conclusions:

• The proposed learning-enhanced SLIP model effectively improves robotic legged locomotion.
• Adaptive control and morphology are crucial for achieving robust and stable robotic movement.
• Trial-and-error learning provides a viable pathway for robots to master complex locomotion tasks.