Perception Perturbation Detection Using Cross-Model Consistency

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Attackers can induce perturbations in sensor data, leading to inaccurate object detection by perception systems, posing safety risks in autonomous systems.

Innovation Solution

Implement consistency-based perturbation detection using multiple models (e.g., object detection and segmentation models) to calculate a consistency score, identifying perturbations based on a threshold, and localize perturbed pixels within sensor data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If multiple models are used to produce output data for consistency checking, then perturbation detection accuracy is improved, but device complexity increases

Engineering Contradiction:

Improveperturbation detection accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system divides the perturbation detection task into separate functional components: a first model for generating initial output data, a second model for generating alternative output data, and a consistency scoring mechanism for comparison. This segmentation allows each component to specialize in specific aspects of detection, improving overall accuracy while maintaining manageable complexity through modular architecture.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The consistency score acts as an intermediary mechanism that mediates between the outputs of multiple models and the final perturbation detection decision. Instead of directly comparing model outputs or using complex ensemble methods, the system uses consistency scoring as an intermediate step to determine whether perturbations are present, simplifying the decision-making process while maintaining high detection accuracy.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If consistency score threshold is set low to detect more perturbations, then detection sensitivity is improved, but false positive rate increases

Engineering Contradiction:

Improvedetection sensitivityVSAvoidfalse positive rate

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The system incorporates feedback mechanisms where the consistency score is continuously compared against the threshold, and the detection results feed back into the system's understanding of perturbation patterns. This feedback loop allows the system to learn from previous detections and adjust its sensitivity, reducing false positives while maintaining high detection sensitivity for actual perturbations.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system dynamically adjusts the consistency score threshold parameter based on operational context and performance metrics. Rather than using a fixed threshold, the system can modify the threshold parameter to optimize the balance between detection sensitivity and false positive rate for different scenarios, allowing adaptive control of detection behavior.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20260073701A1Consistency-based perturbation detection for perception systems

Publication Date: 2026.03.12 QUALCOMM INC

AI summary

Systems and techniques are described herein for perturbation detection. For example, a computing device can produce, using a first model based on input data, first output data. The computing device can produce, using a second model based on the input data, second output data. The computing device can determine, based on the first output data and the second output data, a consistency score indicating a consistency between the first output data and the second output data. The computing device can determine, based on the consistency score being less than a consistency score threshold, the input data comprises a perturbation.