Standing Balance Control of a Bipedal Robot Based on Behavior Cloning.

Area of Science:

• Robotics
• Machine Learning
• Control Systems

Background:

• Bipedal robots offer human-like mobility for versatile applications.
• Controlling balance during physical interaction remains a significant challenge for bipedal robots due to inherent instability.

Purpose of the Study:

• To develop a novel balance controller for bipedal robots utilizing a behavior cloning approach.
• To enhance the stability and dynamic interaction capabilities of bipedal robots in human-scale environments.

Main Methods:

• A behavior cloning model, incorporating two deep neural networks (DNNs), was trained on human-operated balancing data.
• The model predicts the required wrench for balance, with joint torques calculated using robot dynamics.
• Validation was performed on a bipedal lower-body robotic system through simulations and experiments with frontal plane perturbations.

Main Results:

• The proposed balance controller demonstrated superior performance in maintaining balance against perturbations compared to conventional methods.
• The controller generated smoother balancing movements, indicating improved dynamic control.
• The system successfully adapted to external disturbances, showcasing robustness.

Conclusions:

• Behavior cloning offers a promising machine learning-based solution for bipedal robot balance control.
• The developed controller significantly improves stability and control during physical interactions.
• This approach paves the way for more capable and adaptable bipedal robots in complex environments.