Haptic Exploration of Unknown Objects for Robust In-Hand Manipulation.

Area of Science:

• Robotics
• Artificial Intelligence
• Haptics

Background:

• Robotic manipulation in unstructured environments requires knowledge of object properties and motion, which is often unavailable.
• Grasping unknown objects without prior information leads to poor manipulation stability, accuracy, and repeatability.
• Efficiently acquiring object information is a key challenge for in-hand manipulation.

Purpose of the Study:

• To develop an in-hand manipulation framework for unknown objects without prior information.
• To extract object properties using a novel haptic exploration procedure.
• To learn object motion from demonstration using dynamical movement primitives.

Main Methods:

• A human-like robot hand was used for in-hand manipulation experiments.
• A novel haptic exploration procedure was employed to gather object property data.
• Dynamical movement primitives were utilized for learning manipulation trajectories from demonstrations.

Main Results:

• The proposed framework successfully manipulated unknown objects.
• Haptic exploration significantly enhanced manipulation robustness and accuracy.
• Performance was compared against the widely used virtual spring framework baseline.

Conclusions:

• Haptic exploration is an effective method for acquiring object properties in robotic manipulation.
• The developed framework improves the performance of in-hand manipulation for unknown objects.
• This approach offers a pathway for more adaptable and precise robotic manipulation in real-world scenarios.