Grip Stabilization of Novel Objects Using Slip Prediction.

Area of Science:

• Robotics
• Artificial Intelligence
• Materials Science

Background:

• Robotic grasping and in-hand manipulation face challenges with unknown objects due to the vast number of possible interactions.
• Traditional control methods require accurate object models, which are impractical for diverse real-world scenarios.

Purpose of the Study:

• To develop a grip stabilization approach for novel objects that predicts slip onset.
• To enable robots to control contact interactions without explicit object model knowledge.

Main Methods:

• Utilized tactile information (pressure, deformation) from a BioTac sensor to predict slip.
• Formulated a supervised learning problem for slip prediction from high-dimensional tactile data.
• Evaluated the impact of input features, prediction time horizons, and tactile channels on accuracy.

Main Results:

• The slip prediction model generalized to objects not encountered during training.
• The approach successfully stabilized previously unknown objects by counteracting slip events.
• Controller feedback loop incorporating slip prediction demonstrated effective grip stabilization.

Conclusions:

• Slip prediction from tactile sensing offers a viable alternative to model-based control for robotic grasping.
• This method enhances robotic manipulation capabilities by enabling stable handling of diverse, unknown objects.
• The developed approach contributes to more adaptable and robust robotic systems in unstructured environments.