导航模型分布式训练方法、装置、计算机设备及存储介质
By employing a distributed training method for navigation models, the problems of the inability of intelligent agent navigation models to be personalized and the leakage of privacy data are solved, achieving efficient adaptation to the intelligent agent environment and protection of data privacy.
CN119066428BActive Publication Date: 2026-07-17PING AN TECH (SHENZHEN) CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PING AN TECH (SHENZHEN) CO LTD
- Filing Date
- 2024-09-04
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing intelligent agent navigation models are trained on a single server, which makes it impossible to personalize and adapt to the intelligent agent environment and poses a risk of privacy data leakage.
Method used
A distributed training method for navigation models is adopted. An initial encoding model is trained by acquiring an environmental image library and natural language instructions, a global navigation model is generated and distributed to an agent cluster, and the agents update the training parameters based on local data, and finally generate the target navigation model.
Benefits of technology
It improves the intelligence and adaptability of navigation models, avoids privacy data leakage, and enhances the ease of monitoring and debugging of model updates.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN119066428B_ABST
Abstract
本申请涉及机器视觉领域，公开了一种导航模型分布式训练方法、装置、设备及介质，方法包括：基于环境图像与自然语言指令对初始编码模型进行训练得到目标编码模型；基于目标编码模型生成用于根据环境图像生成动作策略的全局导航模型；将全局导航模型分别下发至智能体集群的每一智能体，以使每一智能体基于本地数据对全局导航模型进行模型参数训练得到第一训练参数；获取各智能体反馈的第一训练参数以对全局导航模型进行参数更新以生成目标导航模型，因此本申请无需将智能体的本地数据集中中心节点，确保智能体本地数据的隐私安全，可以结合智能体所处环境，以学习更多的本地化特征和场景，适应不同的环境和任务需求。
Need to check novelty before this filing date? Find Prior Art