Neural Network Grasp Control for Accurate Sliding Manipulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing grasping apparatuses struggle to perform work involving sliding motions with high accuracy and efficiency, especially when the target object changes, leading to increased calculation and working time when using feedback control and pre-programmed operations.
Innovation Solution
An arithmetic device that includes an acquisition unit for state variables, a storage unit with a learned machine learner, and an arithmetic unit to calculate and output target values for actuators, enabling real-time optimization of sliding motions by using machine learning with supervised learning techniques involving sensors like image, rotation, force, vibration, and audio data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If feedback control and pre-programmed operations are used to increase work accuracy, then manufacturing precision is improved, but loss of time increases due to increased calculation requirements
Solution Approach 1:
The system performs preliminary learning by collecting training data during actual work execution, storing the relationship between state variables and work quality in a database. This pre-computed knowledge is then reused during actual operations, eliminating the need for real-time complex calculations while maintaining high accuracy.
Solution Approach 2:
The system creates a virtual model by copying and storing the relationship between state variables and work quality outcomes in a database. Instead of performing complex real-time calculations, the system queries this pre-stored knowledge base to determine optimal actuator target values, significantly reducing computation time while maintaining accuracy.
2Manufacturing precision
If complex feedback control calculations are performed in real-time, then manufacturing precision is improved, but productivity decreases due to increased calculation time
Solution Approach 1:
The system performs the computationally intensive learning process in advance by collecting training data during work execution and storing the results in a database. During actual production, the system only needs to query this pre-computed knowledge, dramatically reducing real-time calculation requirements and improving productivity while maintaining precision.
Solution Approach 2:
The system replaces real-time mechanical feedback control calculations with a database query-based approach. By substituting complex real-time computation with pre-computed knowledge retrieval, the system achieves both high precision and improved productivity.
3Ease of operation
If the robot is driven by a pre-prepared operation program, then ease of operation is improved, but adaptability decreases when target objects change
Solution Approach 1:
The system incorporates feedback by continuously monitoring state variables during work execution and using this information to query the database for optimal actuator target values. This feedback mechanism enables the system to adapt to different target objects while maintaining ease of operation through automated adjustment.
Solution Approach 2:
The system transitions from static pre-programmed operations to dynamic adaptive control. By using the database of learned relationships and querying it based on real-time state variables, the system can dynamically adjust its behavior to accommodate different target objects while maintaining operational simplicity.
Data Source
AI summary
The arithmetic device configured to perform a calculation for controlling a motion of a grasping apparatus that performs work involving a motion of sliding a grasped object includes: an acquisition unit configured to acquire a state variable indicating a state of the grasping apparatus during the work; a storage unit storing a learned neural network that has been learned by receiving a plurality of training data sets composed of a combination of the state variable acquired in advance and correct answer data corresponding to the state variable; an arithmetic unit configured to calculate a target value of each of various actuators related to the work of the grasping apparatus by inputting the state variable to the learned neural network read from the storage unit; and an output unit configured to output the target value of each of the various actuators to the grasping apparatus.