Human-taught sensory-control synergy for universal robotic grasping

Area of Science:

• Robotics
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Universal grasping is crucial for robotic manipulation, especially for humanoid robots.
• Current methods face challenges in achieving human-like dexterity and adaptability.

Purpose of the Study:

• To develop a novel sensory-control synergy (SCS) approach for universal robotic grasping.
• To mimic human grasping experiences using multimodal tactile data and neurocognitive principles.

Main Methods:

• A tactile glove captures multimodal data (contact, slip, pressure) during human demonstrations.
• Neural networks encode tactile data into semantic grasping states.
• An experience-based fuzzy controller translates states into robotic actions.

Main Results:

• The SCS model demonstrates high logicalization and generalizability.
• Achieved a 95.2% success rate in grasping diverse objects (slippery, fragile, soft, heavy).
• Robots showed adaptive manipulation in dynamic disturbances and complex tasks.

Conclusions:

• The SCS approach enables data-efficient learning and transferability to robots.
• The system exhibits human-like universal grasping capabilities.
• This method advances dexterous manipulation in humanoid robots.