Radar Gesture Detection Training for Low-Power Ambient Interaction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing smart devices with physical user interfaces require attention and are inefficient for interaction, and ambient computing faces challenges such as power consumption, environmental variations, background noise, size constraints, and user privacy.

Innovation Solution

A radar-based system with a machine-learned module for gesture detection that includes a two-phase evaluation process, using pre-segmented and unsegmented recognition tasks to filter noise and reduce false positives, while consuming low power and being resistant to environmental changes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If a physical user interface is used for smart device interaction, then the device can be operated reliably, but the user must devote attention away from primary tasks, making interaction cumbersome and inefficient

Engineering Contradiction:
ImproveInteraction efficiencyVSAvoidTime spent on device interaction
Core Design Contradiction:
Ease of operationVSLoss of time

Solution Approach 1:

The patent replaces the mechanical touch interface with a radar-based gesture recognition system. The radar system detects gestures through electromagnetic waves and uses machine learning to classify them, eliminating the need for physical contact and visual attention to the device interface.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent introduces an intermediary machine learning module that processes radar data and translates gestures into device commands. This intermediary layer enables natural, eye-free interaction by interpreting user intentions without requiring direct device contact or visual engagement.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Ease of operation

If a radar system is implemented for ambient computing, then eye-free interaction and reduced cognitive load are achieved, but power consumption increases

Engineering Contradiction:
ImproveEye-free interaction capabilityVSAvoidPower consumption
Core Design Contradiction:
Ease of operationVSUse of energy by moving object

Solution Approach 1:

The patent employs periodic radar signal transmission with duty cycling, where the radar operates in intervals rather than continuously. This periodic operation maintains gesture detection capability while significantly reducing average power consumption compared to continuous operation.

Inventive Principle:
Principle #19Periodic action

Solution Approach 2:

The machine learning module is trained to efficiently distinguish genuine gestures from background noise, enabling the system to self-regulate its processing intensity. This reduces unnecessary computational power consumption while maintaining accurate gesture recognition.

Inventive Principle:
Principle #25Self-service

3Reliability

If gesture recognition is performed in ambient environments, then user privacy is maintained, but background noise and environmental variations increase false positive rates

Engineering Contradiction:
ImproveUser privacy protectionVSAvoidBackground noise interference
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent performs preliminary training of the machine learning module using diverse ambient data that includes various background noises and environmental variations. This preliminary exposure enables the model to learn robust gesture patterns and filter out environmental interference during actual operation.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent adjusts classification parameters and thresholds based on environmental conditions and noise levels. By dynamically changing decision parameters, the system maintains high accuracy in distinguishing genuine gestures from background noise across varying ambient conditions.

Inventive Principle:
Principle #35Parameter changes

4Speed

If unsegmented continuous time-series data is used for gesture recognition, then real-time detection is achieved, but the recognition task becomes significantly more challenging with higher error rates

Engineering Contradiction:
ImproveReal-time gesture detectionVSAvoidGesture classification accuracy
Core Design Contradiction:
SpeedVSMeasurement precision

Solution Approach 1:

The patent segments continuous time-series radar data into discrete gesture candidates based on motion patterns and temporal characteristics. This segmentation transforms the challenging continuous recognition problem into manageable discrete classification tasks while maintaining real-time detection capability.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system uses feedback from the machine learning classification results to adjust segmentation parameters and refine gesture candidate identification. This iterative feedback loop improves classification accuracy over time while maintaining real-time processing speed.

Inventive Principle:
Principle #23Feedback

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

Enables efficient, eye-free interaction with smart devices at a distance, reducing cognitive load and maintaining privacy, with low power consumption and robustness across varying environments.

Implementation Method 1

The radar system uses an ambient-computing machine-learned module to quickly recognize gestures performed by a user up to at least two meters away

Methodology Applied
Scientific EffectRadar: Radar

Data Source

PatentEP4320552B1Training machine-learned modules for radar-based gesture detection in an ambient computer environment
Publication Date: 2025.07.02 GOOGLE LLC
  • EP4320552B1 patent drawingFigure 1-1
  • EP4320552B1 patent drawingFigure 1-2
  • EP4320552B1 patent drawingFigure 1-3

AI summary

Techniques and apparatuses are described that train machine-learned modules to perform radar-based gesture detection in an ambient compute environment. Compared to other smart devices that rely on a physical user interface, a smart device (104) with a radar system (102) can support ambient computing by providing an eye-free interaction and less cognitively demanding gesture-based user interface. The radar system (102) can be designed to address a variety of challenges associated with ambient computing, including power consumption, environmental variations, background noise, size, and user privacy. The radar system (102) uses an ambient- computing machine-learned module (222) to quickly recognize gestures performed by a user up to at least two meters away. The ambient-computing machine-learned module (222) is trained, at least in part, using a two-phase evaluation process, which includes a segmented classification task and an unsegmented recognition task.