Method and apparatus for providing skills for actions of agent in computing system, and skill learning method

A data-based policy adaptation using a large language model and diffusion planner allows zero-shot adaptation to natural language contexts, addressing limitations in existing technologies and improving real-time responsiveness in AI applications.

WO2026127338A1PCT designated stage Publication Date: 2026-06-18RES & BUSINESS FOUND SUNGKYUNKWAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
RES & BUSINESS FOUND SUNGKYUNKWAN UNIV
Filing Date
2025-10-20
Publication Date
2026-06-18

AI Technical Summary

⚠Technical Problem

Existing technologies face limitations in adapting policies zero-shot to new contexts specified in natural language, such as changing environmental conditions or various user requirements, without environmental interactions or expert data sets.

⚗Method used

A data-based policy adaptation technology using a large language model (LLM) to adapt an agent's policy in a zero-shot manner by generating a skill trajectory aligned with a natural language context through a diffusion planner.

🎯Benefits of technology

Enables real-time policy adaptation in response to user instructions without additional learning, enhancing performance in fields like robot operation and automation systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure KR2025016588_18062026_PF_FP_ABST
    Figure KR2025016588_18062026_PF_FP_ABST
Patent Text Reader

Abstract

The present invention relates to a method for providing skills for actions of an agent of a computing system, comprising the steps of: acquiring a natural language-based context; adapting, in a zero-shot manner, a policy of the agent from the context by using a diffusion planner trained to generate a skill trajectory corresponding to the context; and applying skills in the skill trajectory based on the policy to the agent by reflecting the skill trajectory based on the policy to the diffusion planner.
Need to check novelty before this filing date? Find Prior Art