Quickly scalable full-chain simulation operational data generation method and system

By using large language models and cross-instance proximity alignment technology, robot operation data is automatically generated in a simulation environment. This solves the problems of data generation relying on human labor and the lack of diversity in existing technologies, and realizes rapid and scalable full-chain data generation, which is suitable for a variety of robot application scenarios.

CN122154383APending Publication Date: 2026-06-05SHANDONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SHANDONG UNIV
Filing Date
2026-01-07
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing methods for generating robot operation data rely on human labor, making it difficult to scale up and generate data that is singular. In particular, it is difficult to generate full-chain data in complex tasks and diverse scenarios, resulting in low data collection efficiency and limited coverage.

Method used

A large language model is used to generate operation task descriptions. Combined with cross-instance proximity alignment technology, a scenario is constructed in the simulation environment and the target state is transferred to achieve automated generation of operation data across the entire chain.

Benefits of technology

It enables large-scale data generation without human intervention. The generated data covers task descriptions, scene configurations, and status information, and has stronger generalization capabilities and practical value, making it suitable for various robot application scenarios such as home services and industrial operations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122154383A_ABST
    Figure CN122154383A_ABST
Patent Text Reader

Abstract

The application belongs to the technical field of robot training data generation, and discloses a full-chain simulation operation data generation method and system which can be quickly scaled, comprising the following steps: a database containing different kinds of object models is constructed; based on a large language model, a plurality of operation task descriptions are generated by performing operation tasks on the operable objects in the database; according to each operation task description, a corresponding operation scene layout is generated and a simulation scene is constructed in a simulation environment; in the simulation scene of each task, a target state demonstration is collected through teleoperation, and the target state is migrated to other simulation scenes under the same task by using a cross-instance near-touch alignment technology to generate a corresponding operation target state; and finally, the task descriptions, rendered images, depth information, initial states and target states in all simulation scenes are integrated to construct a large-scale robot operation dataset. The application solves the problems of existing data generation methods, such as dependence on manpower, difficulty in scaling and single data.
Need to check novelty before this filing date? Find Prior Art