defense 2026

Detection of Adversarial Attacks in Robotic Perception

Ziad Sharawy , Mohammad Nakshbandiand , Sorin Mihai Grigorescu

Transylvania University of Brașov

0 citations
α

Published on arXiv

2603.28594

Input Manipulation Attack

OWASP ML Top 10 — ML01

Deep Neural Networks (DNNs) achieve strong performance in semantic segmentation for robotic perception but remain vulnerable to adversarial attacks, threatening safety-critical applications. While robustness has been studied for image classification, semantic segmentation in robotic contexts requires specialized architectures and detection strategies.

Key Contributions

  • Framework for detecting adversarial attacks using pre-trained ResNet-18 and ResNet-50 for semantic segmentation
  • Statistical detection metrics combining confidence scores, non-maximal entropy, and kernel density for distinguishing clean from adversarial inputs
  • Comparative analysis of network architectures to identify factors enhancing robustness in robotic perception

🛡️ Threat Analysis

Input Manipulation Attack

Paper focuses on detecting adversarial examples targeting semantic segmentation models at inference time - adversarial inputs crafted to cause misclassification in robotic perception.

Details

Domains
vision
Model Types
cnn
Threat Tags
inference_time
Applications
semantic segmentationrobotic perceptionautonomous driving
Read PDF arXiv Code

Similar Papers

defense

Counterexample Guided Branching via Directional Relaxation Analysis in Complete Neural Network Verification

2026 0 cit.
Input Manipulation Attack
89%
defense

Lipschitz-aware Linearity Grafting for Certified Robustness

2025 0 cit.
Input Manipulation Attack
89%
defense

CIARD: Cyclic Iterative Adversarial Robustness Distillation

2025 0 cit.
Input Manipulation Attack
89%
defense

Algebraic Robustness Verification of Neural Networks

2026 0 cit.
Input Manipulation Attack
89%
defense

Entropy-Based Non-Invasive Reliability Monitoring of Convolutional Neural Networks

2025 0 cit.
Input Manipulation Attack
89%
defense

Knowledge-Guided Adversarial Training for Infrared Object Detection via Thermal Radiation Modeling

2026 0 cit.
Input Manipulation Attack
89%
defense

Diffusion-Based Feature Denoising and Using NNMF for Robust Brain Tumor Classification

2026 0 cit.
Input Manipulation Attack
89%
defense

Why the Maximum Second Derivative of Activations Matters for Adversarial Robustness

2026 0 cit.
Input Manipulation Attack
89%