Co-optimization of robotic design and skill inspired by human hand evolution

Area of Science:

• Robotics
• Evolutionary Computation
• Mechanical Engineering

Background:

• Human hand evolution is influenced by learned skills and environmental interactions.
• Underactuated robotic hands offer adaptive grasping but require optimized mechanical parameters.

Purpose of the Study:

• To propose a co-optimization approach for underactuated robotic hands.
• To integrate grasping skills and structural parameter optimization.
• To enhance robotic hand dexterity and adaptability.

Main Methods:

• A co-optimization framework parameterizes hand design and updates mechanical parameters globally.
• A 'grasping feeling' mechanism estimates stability to accelerate training.
• The method is applied to optimize torsion spring parameters in a multi-actuator underactuated robotic hand.

Main Results:

• Optimized torsion spring parameters improved the physical robotic hand's grasping performance.
• The collaborative evolution process demonstrated enhanced dexterity and adaptivity.
• The approach successfully transferred optimized parameters to a physical system.

Conclusions:

• The proposed co-optimization method effectively enhances underactuated robotic hand grasping.
• Integrating learned skills with mechanical design offers a promising direction for robotic hand development.
• This approach leverages multi-actuator dexterity and underactuated mechanism adaptivity.