Robot Motion Parameter Approximation for Simulator Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies require a large amount of movement data and increase calculation costs when converting control command values using deep neural networks to align simulator models with real robots, due to differences in motion characteristics.
Innovation Solution
An information processing device and method that approximates motion characteristic parameters of a simulator model based on the movement results of a real robot, using similarity between simulation results with different motion characteristic parameters, allowing for efficient and automatic adaptation without continuous recalibration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If control command values are converted using a deep neural network to align simulator models with real robots, then motion characteristic differences are absorbed, but calculation costs increase
Solution Approach 1:
The patent pre-calculates and stores correspondence relationships between simulator output values and real robot input values in a lookup table before actual control execution. This preliminary action eliminates the need for real-time deep neural network calculations, significantly reducing computational costs while maintaining motion characteristic alignment accuracy.
Solution Approach 2:
The patent creates a simplified lookup table that copies and stores the essential mapping relationships between simulator and real robot parameters. Instead of using a complex deep neural network model, this copied representation maintains the necessary correspondence information in a lightweight format that enables fast query-based conversion.
2Measurement precision
If deep neural network conversion is used to align motion characteristics, then accuracy is improved, but the complexity of the system increases
Solution Approach 1:
The patent replaces the complex deep neural network model with a simplified lookup table that copies the essential mapping relationships. This copying approach maintains accuracy by preserving the correspondence data while dramatically reducing system complexity and removing the need for complex inference computations.
Solution Approach 2:
The patent uses a lightweight lookup table structure that is computationally inexpensive and can be easily stored in memory. This disposable-like structure replaces the heavy, complex neural network model with a simple data structure that requires minimal processing resources while achieving the same alignment function.
3Reliability
If a large amount of movement data is collected for training, then the motion characteristic alignment is improved, but the time and resources required increase
Solution Approach 1:
The patent performs data collection and correspondence relationship establishment as a preliminary offline training phase. During this preliminary action, movement data is collected and processed to build the lookup table, which is then stored for reuse. This separates the time-consuming data processing from real-time operation, reducing training time impact on deployment.
Solution Approach 2:
The patent extracts and copies only the essential correspondence relationships from the training data into a compact lookup table format. Instead of storing or processing the entire large dataset during operation, it copies the distilled mapping information, maintaining alignment accuracy while reducing the time and resources needed for actual use.
Data Source
AI summary
[Problem] To more easily and effectively absorb a difference in motion characteristic between a simulator model and a real robot.[Solution] Provided is an information processing device including a communication unit that receives a movement result of an autonomous moving body based on a control command value, and a parameter approximation unit that approximates a motion characteristic parameter of a simulator model for a movement simulation of the autonomous moving body on the basis of a movement result of the autonomous moving body, in which the parameter approximation unit approximates the motion characteristic parameter on the basis of similarity between a plurality of simulation results acquired on the basis of the different motion characteristic parameters in the movement simulation based on the control command value, and the movement result of the autonomous moving body.