Online gray level updating and safe rollback method for robot end side model

By aligning multi-source data in time sequence and learning safe transition kernel functions online, smooth mapping and safe rollback of the robot end-side model are achieved, solving the problems of numerical mutations and control command continuity during model switching and ensuring safe rollback of the system in case of anomalies.

CN122450489AActive Publication Date: 2026-07-24HINTON ARTIFICIAL INTELLIGENCE TECHNOLOGY CO LTD
6 Cites 0 Cited by

Patent Information

Application Number
CN202610856859.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-06-15
Publication Date
2026-07-24
Estimated Expiration
2046-06-15

AI Technical Summary

Technical Problem

Existing robot end-side model update schemes suffer from numerical step jumps and control command continuity issues during model switching when dealing with highly dynamic physical systems. Furthermore, the rollback mechanism fails to effectively integrate with physical constraint boundaries, resulting in a lack of smooth transition and safety assurance in abnormal situations.

Method used

By employing multi-source data time-series alignment and fusion, differential encoding of actions between old and new models, online learning of safe transition kernel functions, and hard real-time dynamic envelope verification, we can achieve smooth mapping of instructions between old and new models and safe rollback under abnormal conditions.

Benefits of technology

By eliminating heterogeneity of multi-source data through temporal alignment and learning safe transition kernel functions online using action difference features, the continuity and safety of the model update process are ensured, and abnormal data is prevented from spreading to the physical execution layer.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses an online gray level updating and safe rollback method for a robot end side model, and belongs to the technical field of robot control and data processing. The method comprises the following steps: acquiring data of a sensor and an actuator to generate a state vector and an executable constraint, performing parallel reasoning on a new model and an old model, and performing differential coding; constructing a local feasible pipeline and a safety transition kernel function, and realizing smooth mapping of model actions; a hard real-time safety monitor is adopted to execute dynamic envelope verification, and the physical safety of the updating process is guaranteed; and realizing stable model takeover based on progress, safety and convergence three-evidence combined judgment, and performing fast rollback through a reverse safety transition kernel function in case of abnormality. According to the method, the problems of action step, discontinuous rollback, insufficient real-time performance and the like easily caused by traditional model switching are solved, non-interruption, low-disturbance and high-safety online updating of the end side model is realized, the method is suitable for real-time control systems such as industrial robots and collaborative robots, and the operation continuity and deployment reliability of equipment are improved.
Need to check novelty before this filing date? Find Prior Art