Trajectory prediction and forecasting of moving objects on roads

Abstract

This project explores the implementation and analysis of advanced SLAM (Simultaneous Localization and Mapping) techniques for autonomous vehicles, focusing on real-time trajectory prediction and forecasting of moving objects. Using the ALIVE vehicle platform equipped with multiple Velodyne LiDAR sensors and Intel RealSense cameras, three cutting-edge SLAM frameworks—KISS-ICP, ORB-SLAM3, and CT-ICP—were integrated and evaluated. Each framework contributes unique strengths: KISS-ICP for lightweight and efficient LiDAR-based mapping, ORB-SLAM3 for robust visual-inertial SLAM, and CT-ICP for continuous trajectory estimation in dynamic environments. The challenges encountered during implementation, such as ROS compatibility issues and parameter tuning, were systematically addressed to achieve robust and reliable localization and mapping performance. This report provides insights into the comparative strengths of these frameworks and highlights their applicability in enhancing autonomous navigation systems for unstructured and dynamic scenarios.