Robot Model Learning With External Force Feedback Control

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Setting parameters and designing a reward function is difficult when learning a robot model by machine learning, making it challenging to efficiently learn the model.

Innovation Solution

A robot model learning device that includes an acquisition unit for actual values, a state transition model and external force model to calculate predicted values, a reward calculation unit to determine optimal action commands, and update units to refine the models based on actual and predicted values, distinguishing between modified and adversarial external forces to enhance learning efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If machine learning is used to learn a robot model, then the robot can automatically acquire control laws, but it becomes difficult to set parameters and design reward functions, reducing learning efficiency

Engineering Contradiction:

Improveautomatic acquisition of control lawsVSAvoiddifficulty in setting parameters and designing reward functions

Core Design Contradiction:

Extent of automationVSDevice complexity

Solution Approach 1:

The system enables the robot model to learn control laws autonomously through self-service machine learning, where the robot automatically acquires control laws without requiring manual parameter setting or reward function design by operators

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system incorporates feedback mechanisms where the robot model continuously learns from execution results and external force information, using this feedback to automatically refine control laws without requiring external intervention for parameter adjustment

Inventive Principle:

Principle #23Feedback

2Reliability

If traditional machine learning approaches are used, then robot control laws can be learned, but the learning process is inefficient due to difficulty in parameter setting and reward function design

Engineering Contradiction:

Improverobot control law acquisitionVSAvoidlearning efficiency

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The system uses feedback from external force information and execution results to guide the learning process, where the robot model automatically adjusts its control laws based on real-time feedback, significantly improving learning efficiency without sacrificing reliability

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system automatically changes and optimizes parameters during the learning process through machine learning algorithms, eliminating the need for manual parameter setting and enabling efficient adaptation to different control scenarios

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12521876B2Robot model learning device, robot model machine learning method, recording medium storing robot model machine learning program, robot control device, robot control method, and recording medium storing robot control program

Publication Date: 2026.01.13 OMRON CORP

AI summary

A non-transitory recording medium storing a robot control device acquires an actual value of a position and a posture of a robot and an actual value of an external force applied to the robot, executes a robot model including a state transition model that, based on the actual value of the position and the posture at a certain time and an action command that can be given to the robot, calculates a predicted value of the position and the posture of the robot, and an external force model that calculates a predicted value of an external force applied to the robot, calculates a reward based on an error of the position and the posture and the predicted value of the external force, generates and gives a plurality of candidates for the action command to the robot model for each control cycle.