Neuromorphic Vision Based Contact-Level Classification in Robotic Grasping Applications.

Area of Science:

• Robotics
• Artificial Intelligence
• Sensor Technology

Background:

• Robotic sorting is increasingly vital in industry, necessitating advanced sensing solutions.
• Conventional vision-based tactile sensing often suffers from high latency and power demands.
• Neuromorphic sensing offers a promising alternative for efficient robotic applications.

Purpose of the Study:

• To introduce a novel neuromorphic vision-based tactile sensing approach for robotic sorting.
• To develop and evaluate Machine Learning (ML) classifiers for object property classification.
• To demonstrate the system's capability for automated object sorting and adaptive grasping.

Main Methods:

• Development of a neuromorphic vision-based tactile sensing system.
• Implementation of Support Vector Machine (SVM) and Dynamic Time Warping-K Nearest Neighbor (DTW-KNN) classifiers.
• Creation of an Event-Based Object Grasping (EBOG) experimental setup for data acquisition (243 experiments).

Main Results:

• The proposed ML methods achieved high prediction accuracy for material hardness, object size, and grasping force.
• Objects were automatically sorted based on classifier predictions, with adaptive re-grasping for low-accuracy predictions.
• The SVM model demonstrated superior accuracy and efficiency compared to the DTW-KNN model for real-time classification.

Conclusions:

• The developed neuromorphic vision-based tactile sensing approach is effective and applicable for robotic sorting.
• ML-based classification enables accurate identification of object properties for automated sorting.
• The SVM classifier provides a robust and efficient solution for real-time tactile sensing in robotic manipulation.