一种基于视觉语言动作模型的灵巧手运动规划与控制方法

By using a visual language motion model, the control bottleneck of dexterous hands in complex scenarios was solved, achieving efficient, stable and smooth motion control with limited data, and improving the generalization ability and reliability of dexterous hands in multimodal human-machine collaborative scenarios.

CN121696972BActive Publication Date: 2026-07-17SHENZHEN RUIYAN INTELLIGENT CONTROL CO LTD

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SHENZHEN RUIYAN INTELLIGENT CONTROL CO LTD
Filing Date
2026-01-19
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Traditional dexterous hands face multiple bottlenecks in terms of data and performance, including high-cost data acquisition, inefficient reinforcement learning, difficulty in generating action sequences for fine control requirements with single-modal models, and motion jitter and error accumulation during the deployment phase, which limit their generalization ability and reliability in complex multimodal human-machine collaborative scenarios.

Method used

A visual-language action model-based approach is adopted. By synchronizing multi-channel camera images and language task commands in a timely manner, combined with dynamic adaptive exponential moving average filtering, feature fusion of visual encoder and language encoder, and a double buffering mechanism to train the model, a smooth control sequence is generated through time fusion, speed limiting and Bezier interpolation. The deployed thread monitors in real time and allows for manual intervention, forming a robust control system.

Benefits of technology

With limited teaching data, the practicality and engineering deployability of the dexterous hand were improved, high-frequency jitter was suppressed, grasping stability and real-time response capabilities were enhanced, robustness to changes in lighting and external disturbances was improved, data acquisition costs and manual annotation requirements were reduced, and operational safety and fault recovery capabilities were enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121696972B_ABST
    Figure CN121696972B_ABST
Patent Text Reader

Abstract

本发明涉及具身智能控制技术领域,公开一种基于视觉语言动作模型的灵巧手运动规划与控制方法,包括S1、采集数据并对多源数据进行时序同步与一致性校验;S2、对关节序列做动态自适应指数移动平均滤波并计算处理,处理结果重构为适配的结构化特征;S3、融合关节状态后输入动作头网络,按预设比例将在线与离线样本拼接成训练批次,判定结果联合优化模型参数;S4、模型推理得目标动作信息,依次处理生成平滑低抖动的控制序列发送给灵巧手执行;S5、部署线程通过环形缓冲与时间戳对齐下发控制指令,超阈值时触发回退轨迹并允许人工介入,人工介入数据写入离线缓冲。本发明能够在极少示教数据集的条件下,实现灵巧手高效、稳定与平滑的运动控制。
Need to check novelty before this filing date? Find Prior Art