Robot Object Search Using Learned Environmental Knowledge

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional robots are inefficient in locating and identifying objects that are not directly in view, often requiring time-consuming operations like SLAM to exhaustively map environments, which consumes resources and can be disruptive.

Innovation Solution

A method using machine learning models, such as convolutional neural networks, to generate posteriori knowledge about object locations, allowing robots to quickly identify potential locations of objects of interest by analyzing visual frames and adjusting their position or pose for optimal viewing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If robots perform exhaustive mapping operations (SLAM) to locate objects, then they can identify objects not directly in view, but they consume excessive resources (power, processing cycles, memory) and time

Engineering Contradiction:
Improveobject location identificationVSAvoidtime to locate object
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary actions by training machine learning models with a posteriori knowledge about object locations in advance. This pre-computed knowledge allows the robot to quickly query likely object locations without performing exhaustive SLAM operations at runtime, thus resolving the contradiction between accurate object location identification and time consumption.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces machine learning models as an intermediary between the robot's vision sensors and the object location determination process. These models act as a mediator that uses learned patterns from training data to predict object locations, replacing the direct but resource-intensive SLAM approach and enabling faster object location with reduced computational burden during operation.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If robots perform exhaustive mapping operations (SLAM) to locate objects, then they can identify objects not directly in view, but they expend excessive resources (power, processing cycles, memory)

Engineering Contradiction:
Improveobject location identificationVSAvoidpower consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system performs preliminary actions by training machine learning models with a posteriori knowledge about object locations in advance. This pre-computed knowledge allows the robot to quickly query likely object locations without performing exhaustive SLAM operations at runtime, thus resolving the contradiction between accurate object location identification and power consumption.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces machine learning models as an intermediary between the robot's vision sensors and the object location determination process. These models act as a mediator that uses learned patterns from training data to predict object locations, replacing the direct but resource-intensive SLAM approach and enabling faster object location with reduced computational burden during operation.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If robots perform exhaustive mapping operations to locate objects, then they can identify objects not directly in view, but the operations are disruptive in home or business environments

Engineering Contradiction:
Improveobject location identificationVSAvoidenvironmental disruption
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The system performs preliminary actions by training machine learning models with a posteriori knowledge about object locations in advance. This pre-computed knowledge allows the robot to quickly query likely object locations without performing exhaustive SLAM operations at runtime, thus resolving the contradiction between accurate object location identification and environmental disruption.

Inventive Principle:
Principle #10Preliminary action

4Adaptability or versatility

If robots use conventional methods to locate objects, then they can find objects anywhere in the environment, but the search process is time-consuming and inefficient

Engineering Contradiction:
Improveobject location coverageVSAvoidobject location speed
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent applies local quality by using machine learning models to identify specific regions of interest in the environment where objects are likely to be located, based on learned patterns from training data. Instead of uniformly searching the entire environment, the system focuses computational resources on localized areas with higher probability of containing the target object, thus improving object location speed while maintaining adaptability.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system changes parameters by using probability thresholds and confidence levels to dynamically adjust the search strategy. When the machine learning model predicts high-probability locations, the robot can directly navigate to those areas; when confidence is lower, it can expand the search area. This parameter-based approach balances object location coverage and speed adaptively.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3720666B1Learning and applying empirical knowledge of environments by robots
Publication Date: 2025.01.29 GOOGLE LLC
  • EP3720666B1 patent drawingFigure 1
  • EP3720666B1 patent drawingFigure 2
  • EP3720666B1 patent drawingFigure 3A~3B

AI summary

Techniques described herein relate to generating a posteriori knowledge about where objects are typically located within environments to improve object location. In various implementations, output from vision sensor(s) of a robot may include visual frame(s) that capture at least a portion of an environment in which a robot operates/will operate. The visual frame(s) may be applied as input across a machine learning model to generate output that identifies potential location(s) of an object of interest. The robot's position/pose may be altered based on the output to relocate one or more of the vision sensors. One or more subsequent visual frames that capture at least a not-previously-captured portion of the environment may be applied as input across the machine learning model to generate subsequent output identifying the object of interest. The robot may perform task(s) that relate to the object of interest.