Enhancing Perception with Tactile Object Recognition in Adaptive Grippers for Human-Robot Interaction.

Area of Science:

• Robotics
• Artificial Intelligence
• Sensor Technology

Background:

• Disaster scenarios present visibility challenges for robots.
• Tactile perception is crucial for robots to distinguish objects, especially human limbs.
• Adaptive grippers with tactile sensors offer enhanced environmental interaction capabilities.

Purpose of the Study:

• To evaluate tactile sensor integration in adaptive grippers for object recognition.
• To assess the performance of deep convolutional neural networks (DCNNs) with flexible tactile sensors.
• To develop a two-level network for object-type and human/non-human classification.

Main Methods:

• Training a DCNN classifier on 15 classes (human body parts, environment objects) using tactile data from semi-rigid and flexible adaptive grippers.
• Comparing classifier performance against rigid configurations and Support Vector Machine (SVM) classifiers.
• Implementing a two-level output network for dual classification tasks.

Main Results:

• Flexible sensors in adaptive grippers yielded more non-null tactels and lower mean pressure, enabling softer grips suitable for physical human-robot interaction (pHRI).
• A semi-rigid adaptive gripper achieved a 95.13% object recognition rate.
• A rigid sensor configuration achieved a higher human/non-human classification accuracy of 98.78%.

Conclusions:

• Adaptive grippers with tactile sensing enhance robotic capabilities in complex environments like disaster sites.
• Semi-rigid adaptive grippers offer a strong balance between object recognition performance and safe interaction.
• Further research may optimize rigid sensor configurations for specific classification tasks like human detection.