Trajectory prediction for autonomous driving using deep learning approach

• Main Author: Wong, Shang Yi
• Other Authors: Lyu Chen
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Mechanical engineering
• Online Access: https://hdl.handle.net/10356/166991
• Institution: Nanyang Technological University

As more self-driving car companies started their real-world testing, using trajectory prediction in the decision-making and planning process of self-driving systems is crucial to enhance the safety and effectiveness of autonomous vehicles. This project proposes a trajectory prediction algorithm for autonomous vehicles using a deep learning framework. The proposed model consists of four parts: a recurrent neural network to encode temporal information of each vehicle, a graph neural network to encode the spatial relationships between the target vehicle and surrounding vehicles, a self-attention mechanism to process and synthesize the important vehicle embeddings and a final multilayer perceptron to decode and generate the future trajectories. The machine learning model is validated on Argoverse, a large-scale trajectory prediction dataset from real-world self-driving vehicles. Ablation studies have shown that the interaction-aware model can outperform maneuver-based models by a large margin. Furthermore, our map-free model performed considerably well from the validation results obtained among other map-dependent trajectory prediction models validated on the benchmark hosted on eval.ai.