面向室内动态场景的生成式大模型机器人操控方法及系统
By combining a slow-fast dual-flow framework with optical flow and kinematic models, the perception and control gaps in robot manipulation in dynamic scenarios are addressed, enabling accurate tracking and real-time adjustment of moving targets and improving the success rate of robot operations in dynamic environments.
CN121962646BActive Publication Date: 2026-07-17CENT SOUTH UNIV +1
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CENT SOUTH UNIV
- Filing Date
- 2026-04-02
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing robot manipulation methods suffer from perception gaps, time delay gaps, and control gaps in dynamic scenarios, making it difficult to accurately predict the position of moving targets and execute actions, leading to grasping/placement failures.
Method used
A slow-fast dual-stream framework is adopted, which enhances visual input through optical flow, predicts future states by combining kinematic models, and performs real-time closed-loop correction in each control cycle to generate and correct action sequences.
Benefits of technology
It improves the observability of dynamic targets, reduces state drift, realizes high-frequency closed-loop control, and enhances execution stability and robustness in dynamic scenarios.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN121962646B_ABST
Abstract
本发明公开了一种面向室内动态场景的生成式大模型机器人操控方法及系统，其中方法包括：当动作缓冲区剩余长度等于推理时延步数时，预测机器人的未来本体状态；获取机器人采集的相邻RGB图像，计算光流场并映射为类RGB图像；分别对当前RGB图像与类RGB图像进行视觉编码，并拼接形成视觉上下文特征；将视觉上下文特征、未来本体状态及输入指令输入慢流宏规划器，生成宏动作块并写入动作缓冲区；在每个控制周期，弹出动作缓冲区头部动作作为宏动作，并与当前RGB图像、本体状态一同输入快流残差细化器，输出动作残差，并据此对宏动作进行修正并执行；循环执行上述过程，直至结束。本发明使机器人在动态操控场景中能稳定完成目标操作。
Need to check novelty before this filing date? Find Prior Art