Generative reasoning method, system, computer device and medium based on reinforcement learning framework

By employing an availability generalization reasoning method based on a reinforcement learning framework, and utilizing a high-precision binocular stereo vision camera and a thought chain reasoning mechanism, the problem of insufficient out-of-domain generalization ability of multimodal large language models is solved, thereby improving the robot's ability to reason about object availability and operational reliability in unstructured environments.

CN122021940BActive Publication Date: 2026-06-23HONG KONG UNIV OF SCI & TECH (GUANGZHOU)
3 Cites 0 Cited by

Patent Information

Application Number
CN202610450847.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-04-08
Publication Date
2026-06-23
Estimated Expiration
2046-04-08

AI Technical Summary

Technical Problem

Existing availability reasoning techniques, when supported by multimodal large language models, suffer from insufficient out-of-domain generalization ability, making it difficult to adapt to complex scenarios and unseen objects in unstructured environments, resulting in insufficient reliability and accuracy of robot operations.

Method used

An availability generalization reasoning method based on a reinforcement learning framework is adopted. By integrating a high-precision binocular stereo vision camera to acquire multimodal environmental data, availability prior knowledge and thought chain reasoning mechanism are introduced, and initial reasoning is performed in combination with a large language model. Through iterative optimization and generalization enhancement processing, the availability region information of the target is output.

Benefits of technology

It improves the reliability and interpretability of robot reasoning in unstructured environments, enhances its adaptability to unseen objects and complex scenes, and enables more accurate and interpretable operations.

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

Abstract

The application relates to the technical field of robot intelligent control, and in particular relates to a generalization reasoning method and system for affordance based on a reinforcement learning framework, computer equipment and a medium; the method comprises the following steps: acquiring multi-modal environment data collected by a perception system and performing pretreatment; based on the pretreated multi-modal environment data, introducing prior knowledge of affordance to construct an input representation of an affordance reasoning task; performing affordance reasoning calculation through a large language model to generate an initial affordance region prediction; based on a reinforcement learning framework, using a thought chain reasoning mechanism to iteratively optimize and enhance the generalization of the initial affordance region prediction, so as to output target affordance region information. In this way, the technical problem of insufficient out-of-domain generalization capability of the prior art under the support of a multi-modal large language model is solved, and the reliability, interpretability and generalization operation capability of the robot in an unstructured environment are improved.
Need to check novelty before this filing date? Find Prior Art