Keywords: interaction modeling, graph neural networks, autonomous driving.
Social interactions between vehicles in traffic must be well-understood in order to develop safe autonomous systems. In semantic environments such as road traffic, ordinary vectorial representation does not capture the rich relationships of the ego vehicle with its neighbors. Graph neural networks (GNN) are ideally suited in such dynamic environments, as they can model changing conditions in the order and number of vehicles. The relationship information is pre-stated, and GNN can derive higher-level information through these relationships.
The goal of this thesis is to model the interactions among autonomous agents in dense traffic using GNN. In this context, the thesis is expected to include an ablation study to identify which input features (such as velocity, distance to other vehicles, etc.) has the maximum effect on prediction and the effects of changing the graph structure on the interactions among the vehicles.
- Review of relevant literature,
- Training and testing different GNN structures to model interactions among vehicles,
- Ablation study on input features,
- Comparison with existing state-of-the-art work.
Pre-requisites: Python(high), Deep Learning (high), and previous experience on GNNs is a big plus
Simulators: (up-to-change) Carla, DriverGym, highway-env
Start: Available immediately
- Graph-Based Spatial-Temporal Convolutional Network for Vehicle Trajectory Prediction in Autonomous Driving, https://arxiv.org/pdf/2109.12764.pdf
- Joint Interaction and Trajectory Prediction for Autonomous Driving using Graph Neural Networks, https://arxiv.org/pdf/1912.07882.pdf
- CRAT-Pred: Vehicle Trajectory Prediction with Crystal Graph Convolutional Neural Networks and Multi-Head Self-Attention, https://arxiv.org/pdf/2202.04488.pdf