Overview:
The goal of this project is to enhance motion planning in autonomous driving by leveraging large-scale transformer models on publicly available autonomous driving datasets and testing the results on an existing open simulation platform. The focus is on developing efficient, safe, and computationally feasible path planning algorithms in common road scenarios, using data-driven approaches.
The student will focus specifically on trajectory prediction for urban environments using large-scale transformer models. The aim is to enhance prediction accuracy and smooth transitions between dynamic elements like pedestrians, cyclists, and vehicles.
Key Elements:
Data Source: Use large-scale open datasets for training and validation.
Simulation Platform: Implement and evaluate the methods on a public simulation platform.
Focus Area: Concentrate on predicting near-future trajectories of dynamic agents and ensuring safe trajectory planning for the ego-vehicle.
Outcome:
1. A fine-tuned transformer model specifically designed for urban trajectory prediction.
2. An enhanced motion planning module tested and validated on a public simulation platform.
Summary:
The project will be carried out in four work packages (WP). In WP 1, the student will study large-scale autonomous driving datasets, familiarize themselves with the simulation platform, and set up the environment for training and testing the transformer model, resulting in preprocessed datasets and a working simulation environment. In WP 2, the focus will be on fine-tuning pre-trained transformer models using urban driving data, specifically training them for trajectory planning in complex urban scenarios like intersections and pedestrian crossings, with the deliverable being a fine-tuned transformer model. WP 3 involves developing a motion planning module that integrates the trajectory predictions for the ego vehicle, which will be tested in a simulation platform under different traffic conditions and multi-agent interactions. The outcome will be a motion planner that adapts dynamically to the environment. Finally, in WP 4, the student will measure the performance of the developed system in terms of accuracy, safety, and computational efficiency, conducting simulations to assess robustness and compiling the results into a comprehensive report and presentation showcasing improvements in motion planning efficiency.
