Back to ASAM International Conference 2019

Current Trends in Machine Learning for Automated Driving

Dr. Alexander Jungmann
Team Manager “Intelligent Driving Functions”
IAV GmbH 

Automated Driving (AD) is still subject of intensive research and development. In order to increase the level of driving automation, AD systems have to become more intelligent to be able to handle more complex situations without human intervention. In our opinion, Machine Learning (ML) techniques are a major building block in order to overcome these challenges.

An AD system can be divided into perception, planning, and acting. In the perception domain, Deep Learning (DL) techniques are dominating traditional algorithms in terms of detection rate. In the planning domain, Reinforcement Learning (RL) techniques enable AD systems to learn how to make decisions by interacting with the environment.

An opposite approach is end-to-end learning. The entire AD system is considered a pre- trained black box. In fact, Deep Reinforcement Learning (DRL) techniques – a combination of DL and RL - replace the perception, planning, and acting components - or at least major parts of them.

In this talk, we want to discuss the current ML trends for automated driving. While DL techniques are already well established for perception tasks, the application of RL or even DLR techniques for intelligent, automated decision-making is still under investigation.