Scene Representation Learning for Decision-Making in Autonomous Driving Systems using Graph Neural Networks

Navigating a self-driving car in dense urban scenarios is a multi-agent decision-making problem and understanding the scene and agents around the car and also the interaction between them is very important for decision-making. One solution to do this scene understanding is to use Graph Neural Networks (GNNs) and create an embedded vector for the decision-making module. After the scene encoding stage, we will use Reinforcement Learning (RL) to use the embedded vector and generate high-level commands for the control module.

Keywords: robotics, autonomous driving, decision-making, reinforcement learning, graph neural network.

Supervisor: Prof. Ville Kyrki (ville.kyrki@aalto.fi)
Advisor: Eshagh Kargar (eshagh.kargar@aalto.fi)

The intersection-v0 environment from highway-env.

Deliverables

  • Implementing and adapting the code available for a method presented in the paper “Learning Transferable Cooperative Behavior in Multi-Agent Teams” for the “highway-env” simulator.

Practical Information

Pre-requisites: Deep Learning, Reinforcement Learning, Autonomous Driving.
Tools: Python, PyTorch
Start: Available immediately

References

  • Agarwal, A., Kumar, S., & Sycara, K. (2019). Learning transferable cooperative behavior in multi-agent teams. arXiv preprint arXiv:1906.01202.
  • Leurent, E. (2018). An environment for autonomous driving decision-making.