Research on Robot Grasping Based on Deep Learning for Real-Life Scenarios

Area of Science:

• Robotics
• Artificial Intelligence
• Deep Learning

Background:

• Robotic arm applications are traditionally limited to industry due to low intelligence.
• Advancements in deep learning enable the development of highly intelligent robots.
• There is significant potential for robotic arms in everyday scenarios.

Purpose of the Study:

• To investigate multi-object grasping in real-life scenarios using intelligent robots.
• To develop and validate a hybrid deep learning model for predicting robotic grasping strategies.
• To enhance the capabilities of robotic arms for domestic and daily life applications.

Main Methods:

• Theoretical analysis and improvement of convolutional neural networks (CNNs) and residual networks (ResNets).
• Construction of a hybrid grasping strategy prediction model integrating CNNs and ResNets.
• Deployment of the trained model within a robot control system for performance validation.

Main Results:

• The hybrid model demonstrated high prediction accuracy for multi-object grasping strategies.
• The study achieved leading performance in robot grasping success rates.
• The deployed model proved effective in real-life robot control scenarios.

Conclusions:

• The developed hybrid deep learning model significantly advances multi-object grasping capabilities for robots.
• This research paves the way for broader applications of intelligent robotic arms in daily life.
• The findings highlight the potential of combining CNNs and ResNets for sophisticated robotic tasks.