MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641074355 A) filed by Dr. Gunji Sreenivasulu; V Vijayakumar Dasari; Dr. More Praveen; A. Shofia; Hafizuddin Mahmad; Jinkala Swathy; Ramarao Gose; and Dr. G Sreenivasa Reddy on June 16, 2026, for A Novel Adversarial Robustness In Deep Learning Models For Autonomous Vehicle Perception.
Inventors include Dr. Gunji Sreenivasulu; V Vijayakumar Dasari; Dr. More Praveen; A. Shofia; Hafizuddin Mahmad; Jinkala Swathy; Ramarao Gose; and Dr. G Sreenivasa Reddy.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: This invention presents a Novel Adversarial Robustness Framework for Deep Learning Models specifically tailored for Autonomous Vehicle Perception systems, aimed at ensuring reliable and secure object detection, lane recognition, and environmental understanding even under sophisticated adversarial attacks. The proposed system integrates a multi-layered defense mechanism that combines dynamic adversarial training, attention-guided feature purification, and real-time uncertainty estimation using an ensemble of specialized neural networks. By continuously adapting to emerging threat patterns through an intelligent feedback loop, the framework significantly enhances the resilience of perception models without compromising their speed or accuracy in normal driving conditions. This approach delivers superior protection against both digital and physical adversarial perturbations, making autonomous vehicles safer and more trustworthy in real-world scenarios where malicious inputs could otherwise lead to critical failures.
Disclaimer: Curated by HT Syndication.
