Grasping Force Control of Multi-Fingered Robotic Hands through Tactile Sensing for Object Stabilization

Area of Science:

• Robotics
• Artificial Intelligence
• Sensor Technology

Background:

• Grasping force control is crucial for multi-fingered robotic hands to maintain object stability.
• Human dexterity relies on rapid tactile feedback for adjusting grip and reacting to instability.
• Current robotic tactile sensing for grasping force control remains an underexplored area.

Purpose of the Study:

• To develop and evaluate a tactile sensing framework for multi-fingered robotic hands to adjust grasping force for stabilizing unknown objects.
• To enable robotic hands to manipulate objects without prior knowledge of their physical properties or shape.
• To enhance the autonomy and adaptability of robotic manipulation systems.

Main Methods:

• An online detection module using Deep Neural Network (DNN) to identify contact events and object material from tactile data.
• A force estimation method employing Gaussian Mixture Models (GMM) to determine contact force and location from tactile sensor readings.
• An object stabilization controller that utilizes tactile sensing results to adjust the robotic hand's contact configuration.

Main Results:

• The DNN module successfully detected contact events and material properties simultaneously from tactile data.
• The GMM-based method accurately estimated contact forces and locations.
• The integrated framework demonstrated effective object stabilization using tactile feedback in real-world experiments with a Shadow Dexterous Hand.

Conclusions:

• The proposed tactile sensing framework enables precise grasping force control for object stabilization in robotic hands.
• The combination of DNN and GMM provides a robust method for interpreting complex tactile data.
• This research advances the capability of robotic hands in handling unknown objects, mimicking human-like tactile sensing and manipulation.