Visual Tactile Estimation for Robot Grasping of Slippery Objects

The patent introduces a model as an intermediary that links visual information and tactile information. This model is trained using paired visual and tactile data, enabling the system to estimate tactile properties (slipperiness, hardness) from visual input with improved accuracy. The model acts as a mediator that captures the complex relationship between visual appearance and tactile characteristics.

The patent creates a virtual copy of tactile information by generating estimated tactile data from visual information through the trained model. This allows the robot to obtain tactile-like information (slipperiness, hardness estimates) without physical contact, effectively copying tactile properties from visual observations.

The patent performs preliminary action by pre-training the model using paired visual and tactile information before actual use. During the training phase, the system learns the relationship between visual appearance and tactile properties by processing大量 paired data. Once trained, the model can be deployed for real-time tactile estimation without requiring complex real-time processing, thus reducing operational complexity.

The patent replaces the need for complex mechanical tactile sensing systems with an information-processing-based model. Instead of requiring sophisticated tactile sensors and mechanical contact mechanisms, the system uses a trained computational model that processes visual information to estimate tactile properties, substituting mechanical complexity with algorithmic processing.