Intellectual Property India Publishes Patent Application for System And Method For Real-Time Adaptive Trajectory Prediction In Autonomous Vehicles Using Dual Swarm-Optimized Attention-Based Graph Neural Networks Over Big Data Streaming Platforms

MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202641088342 A) filed by P Venkata Krishna on July 20, 2026, for System And Method For Real-Time Adaptive Trajectory Prediction In Autonomous Vehicles Using Dual Swarm-Optimized Attention-Based Graph Neural Networks Over Big Data Streaming Platforms.

Inventors include B. Gnana Deepthi; P Venkata Krishna; and Saritha Vankadara.

The application for the patent was published on July 31, 2026, under issue no. 31/2026.

Abstract: The present invention relates to a system and method for real-time processing of high-velocity data streams generated by multiple sensors in autonomous vehicles (AVs) to enable accurate, low-latency motion prediction and decision-making. The invention provides an integrated framework that combines big data streaming technologies (Apache Kafka and Spark Streaming) with a Graph Neural Network incorporating an Attention mechanism (GNN-Attention) for modeling interactions among traffic agents (vehicles, cyclists, pedestrians). A novel dual swarm intelligence optimization layer is employed: Particle Swarm Optimization (PSO) dynamically tunes model hyperparameters and processing parameters (such as batch interval) in response to varying data arrival rates, while Ant Colony Optimization (ACO) adaptively updates the structure and importance weights of the agent interaction graph based on real-time prediction loss and accuracy feedback. The framework is trained and validated using the Lyft Motion Prediction dataset. Experimental results demonstrate significant improvements, including reduction in Average Displacement Error (ADE) from 2.15 m (baseline GNN) to 1.38 m (full proposed configuration), Final Displacement Error (FDE) from 4.80 m to 3.15 m, and inference time from 95 ms to approximately 48 ms, with processing delays in the range of 40-60 ms and end-to-end delays stable between 55-75 ms. The invention addresses critical challenges of latency, computational efficiency, and dynamic adaptability in real-time AV perception and planning systems, thereby contributing to enhanced safety and reliability of autonomous driving operations. The technical effect includes improved real-time responsiveness under variable traffic and sensor data conditions through adaptive graph structure optimization and parameter tuning within a streaming data pipeline.

Disclaimer: Curated by HT Syndication.