Neural Network Control Inference for Flexible Object Target States
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods struggle to effectively train and infer models for flexible object manipulation, as they fail to accurately predict the future state of objects with multiple degrees of freedom.
Innovation Solution
An inferring device that uses a network trained by machine learning to input data about the object's state and time-series control information, predicting future states and outputting new control information to bring the object into a target state.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods are used for flexible object manipulation, then the system structure remains simple, but the ability to accurately predict future states of objects with multiple degrees of freedom deteriorates
Solution Approach 1:
The patent applies dynamics by transitioning from static conventional methods to a dynamic machine learning model that can adaptively predict future states. The neural network model dynamically processes time-series control information and object state data to generate accurate predictions for flexible objects with multiple degrees of freedom, resolving the contradiction between prediction accuracy and model complexity.
Solution Approach 2:
The patent changes parameters by introducing a machine learning model with learnable parameters that can be trained on training data. The model takes object state parameters and control information as inputs and predicts future state parameters, enabling accurate prediction of flexible object manipulation while managing complexity through parameter optimization during training.
2Reliability
If conventional training methods are used, then the training process remains simple, but the ability to effectively train models for flexible object manipulation deteriorates
Solution Approach 1:
The patent applies preliminary action by implementing a dedicated training process that prepares the model before actual use. The training phase uses training data containing object states and control information to pre-train the model parameters, ensuring the model is ready for reliable inference. This separates the complex training process from the simpler inference process, improving reliability while managing complexity through staged implementation.
3Productivity
If static operations are used, then the system remains simple to implement, but the ability to perform real-time inference and control deteriorates
Solution Approach 1:
The patent applies dynamics by implementing a real-time inference capability that processes current object states and control information to predict future states dynamically. The model operates in real-time during the inference phase, enabling productive control of flexible objects while keeping the runtime system relatively simple compared to the training phase.
Data Source
AI summary
To infer dynamic control information on a controlled object. An inferring device includes one or more memories and one or more processors. The one or more processors are configured to: input at least data about a state of a controlled object and time-series control information for controlling the controlled object, into a network trained by machine learning; acquire predicted data about a future state of the controlled object controlled based on the time-series control information via the network into which the data about the state of the controlled object and the time-series control information have been input; and output new time-series control information for controlling the controlled object to bring the future state of the controlled object into a target state based on the predicted data acquired via the network.