Detecting Transitions from Stability to Instability in Robotic Grasping Based on Tactile Perception

Area of Science:

• Robotics
• Artificial Intelligence
• Sensor Technology

Background:

• Robots perform complex load operations, which can lead to dynamic shifts from stable to unstable states.
• Monitoring these transitions is crucial for safe and efficient robotic task execution.

Purpose of the Study:

• To develop a real-time dynamic state sensing network for robots.
• To accurately predict and classify changes in robotic operational stability during load handling.

Main Methods:

• Utilized tactile images from tactile sensors during robot-object interactions.
• Integrated Convolutional Neural Networks (CNNs) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for temporal information.
• Collected a dataset of transitions from stable to unstable states and employed a sliding window approach for frame sampling.

Main Results:

• Achieved real-time temporal sequence prediction with an inference step of 31.84 ms.
• Attained an average classification accuracy of 98.90% in predicting state changes.
• Demonstrated robustness with high accuracy on previously unseen objects.

Conclusions:

• The proposed network effectively senses and predicts dynamic state changes in robots during load operations.
• The system offers a reliable solution for enhancing robotic safety and performance in real-world applications.