Robot Gripping Control Using Vision and Learned Hand Positioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gripping systems for robots face inefficiencies in gripping operations due to the need for precise measurement and recognition of workpieces, which can be time-consuming and require extensive operator training.
Innovation Solution
A gripping system that utilizes a learning device and a neural network to improve the efficiency of gripping operations by acquiring image information and hand position data, constructing models through machine learning, and generating operation commands for the robot.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional distance sensors and 3D CAD model collation are used to measure and recognize workpieces, then measurement precision is improved, but productivity deteriorates due to time-consuming operations
Solution Approach 1:
The patent replaces the mechanical measurement system (distance sensors and 3D CAD model collation) with a vision-based system using imaging sensors and neural networks. The imaging sensor captures images of workpieces, and the neural network automatically recognizes positions and postures, eliminating the need for physical measurement devices and complex computational geometry operations.
Solution Approach 2:
The system enables the robot to autonomously learn and recognize workpiece characteristics through the neural network, eliminating the need for operator training and manual programming. The neural network automatically adapts to different workpiece types and positions, making the system self-adjusting and reducing human intervention.
2Manufacturing precision
If precise measurement and operator training are implemented for gripping operations, then gripping accuracy is improved, but device complexity increases
Solution Approach 1:
The patent replaces complex operator training procedures and manual measurement systems with an automated vision-based neural network system. The imaging sensor and neural network combination provides automatic workpiece recognition and gripping parameter determination, eliminating the need for skilled operators and complex training protocols.
Solution Approach 2:
The neural network is pre-trained with large datasets of workpiece images and gripping outcomes, enabling it to automatically recognize and adapt to new workpiece types without additional training. This preliminary learning phase allows the system to perform accurate gripping operations immediately upon deployment.
3Ease of operation
If traditional gripping methods are used, then ease of operation is maintained, but productivity deteriorates due to time-consuming operations
Solution Approach 1:
The neural network automatically performs workpiece recognition, position determination, and gripping parameter optimization without human intervention. The system self-adjusts to different workpiece types and positions, maintaining ease of operation while dramatically increasing gripping speed through parallel image processing and rapid neural network inference.
Solution Approach 2:
The imaging sensor continuously captures workpiece images, and the neural network continuously processes these images to determine optimal gripping parameters. This continuous operation eliminates idle time between measurements and gripping actions, maintaining high productivity while keeping the interface simple for operators.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A gripping system (1) according to an aspect of the present disclosure includes: a hand (3) that grips a workpiece (W), a robot (2) that supports the hand (3) and changes at least one of a position of the hand (3) and a posture of the hand (3), an image sensor (4) that acquires image information from a viewpoint interlocked with at least one of the position and the posture of the hand (3), a construction module (700) that constructs a model corresponding to at least a part of a process of specifying an operation command of the robot (2) based on the image information acquired by the image sensor (4) and hand position information representing at least one of the position of the hand (3) and the posture of the hand (3), by machine learning based on collection data, an operation module (600) that executes the operation command of the robot (2) based on the image information, the hand position information, and the model, and a robot control module (5) that operates the robot (2) based on the operation command of the robot (2) operated by the operation module (600).