Master Thesis on “Inverse Reinforcement Learning for Driving Behavior Modeling”

Supervisor: Prof. Ville Kyrki (ville.kyrki@aalto.fi)
Advisor: Dr. Gökhan Alcan (gokhan.alcan@aalto.fi), Daulet Baimukashev (daulet.baimukashev@aalto.fi)

Keywords: inverse reinforcement learning, driving behavior modeling, autonomous driving.

Project Description

The most crucial ability that differentiates human drivers from autonomous systems is being able to intuitively consider the behavior of others and make decisions in a split second based on past driving experiences. To increase safety in transportation, autonomous agents should also get this skill and improve their decision-making mechanism.

The goal of this thesis is to formulate driving behavior modeling as an inverse reinforcement learning (IRL) problem and develop efficient methods to solve it. In this context, the thesis is expected to include implementations of different IRL methods and compare them concerning prediction capabilities.

Deliverables

  • Review of relevant literature,
  • Training and testing different IRL methods,
  • Ablation study on network structures and hyperparameters,
  • Comparison of the implementations concerning prediction accuracies.

Practical Information

Pre-requisites: Python(high), Deep Learning (high), and previous experience on IRL is a big plus
Tools: PyTorch
Simulators: (up-to-change) Carla, DriverGym, highway-env
Start: Available immediately

References