Modelling and simulation of the hand grasping using neural networks

Area of Science:

• Robotics
• Artificial Intelligence
• Biomechanics

Background:

• Understanding hand grasping is crucial for developing advanced prosthetic limbs and robotic manipulators.
• Current methods for simulating grasping lack the adaptability to diverse object geometries and interaction dynamics.

Purpose of the Study:

• To model and simulate human hand grasping using artificial neural networks (ANNs).
• To investigate the potential of ANNs in controlling multi-degree-of-freedom prosthetic devices and robotic end-effectors.
• To establish a foundational understanding of grasping coordination and control mechanisms.

Main Methods:

• The human hand was modeled as a black box, with object properties and time sequences as inputs.
• Artificial neural networks were trained using key hand postures for grasping various object shapes and sizes.
• The back-propagation algorithm was employed to optimize network weights for accurate posture prediction.

Main Results:

• The trained neural network demonstrated the ability to reproduce hand grasping postures for objects of different shapes and sizes.
• A single set of neural network weights was sufficient to generalize grasping simulations across varied object parameters.
• Preliminary results indicate successful modeling of grasping dynamics and coordination.

Conclusions:

• Artificial neural networks offer a viable approach for simulating and understanding complex hand grasping actions.
• The developed ANN model shows promise for enhancing the control and dexterity of prosthetic and robotic hands.
• Further research can build upon these findings to create more intuitive and adaptive robotic grasping systems.