Trajectory prediction for autonomous driving using deep learning approach

• Main Author: Zhang, Zihan
• Other Authors: Lyu Chen
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Trajectory prediction; Motion forecasting; Autonomous driving; Deep learning
• Online Access: https://hdl.handle.net/10356/177560
• Institution: Nanyang Technological University

Trajectory prediction is a pivotal challenge for autonomous driving due to the complex interaction within the traffic scenes. In recent years, deep learning has experienced rapid advancement and has proven highly effective in addressing such challenges, significantly enhancing the safety and reliability of autonomous vehicles. These techniques have been extensively applied throughout various stages of autonomous driving development. VectorNet introduces vector representations for traffic scenarios, effectively capturing the complex interactions among all components and setting the foundation for the further motion prediction task. This project explores the development and improvement of trajectory prediction models for autonomous driving using deep learning strategies. It highlights the deployment of an anchor-based trajectory predictor to enhance the methodologies previously established by VectorNet, showing substantial improvements over the original VectorNet model. The study incorporates ablation studies to validate model choices and emphasize component significance. This work demonstrates the feasibility and effectiveness of deep learning approaches for trajectory prediction tasks.