Phd in The Area „imitation and Reinforcement

**Aufgaben**:
The Mercedes-Benz Group AG is one of the most successful automotive companies in the world. Together with Mercedes-Benz AG, the vehicle manufacturer is one of the largest providers of premium and luxury cars and vans.

We are committed to shaping the future of automotive mobility by developing highly automated driving systems for both highway and urban areas. Our development teams in Germany and California work with state-of-the-art technologies to develop innovative and class-leading systems to provide our customers with the best experience possible. To master this challenge, we are looking for energetic and committed PhD students to conduct research within our Vehicle Intelligence Team in Stuttgart.

The focus of this PhD thesis will be to investigate data-driven approaches to autonomously plan vehicle behaviors in urban and highway scenarios. In recent years, many public datasets have been made available by the autonomous driving research community (e.g., Argoverse, Waymo open dataset or the Citiscapes dataset from Mercedes Benz). However, how to make best use of these datasets to improve behavior and motion planning algorithms is still an open research question. Hence the focus of this thesis will be on imitation and reinforcement learning approaches, which can leverage such datasets offline.
The key challenge that will be addressed is linked to the inevitable distributional shift that happens when deploying data-driven planners online. For this, it is expected to survey the latest state of the art in offline Reinforcement Learning (RL) methods. Offline RL belongs to the family of counter-factual inference problems, which given a particular set of decisions, consist of finding better decisions leading for instance to safer and more comfortable driving behaviors. These are known to be especially challenging machine learning problems. You will be expected to make use of the latest machine learning techniques in uncertainty estimation, density estimation and distributional robustness to improve the state of the art in Offline RL and imitation learning for autonomous vehicles.

**Responsibilities
- Assessment of the current state-of-the-art in offline reinforcement and imitation learning.
- Development of imitation and offline reinforcement learning algorithms.
- Definition of metrics that describe the quality of a trajectory based on expert knowledge and recorded human driving data.
- Implementation of a corresponding training and evaluation concept.
- Integration of the developed methods into the AD software.
- Evaluation of the developed methods in simulation, recorded driving data and real vehicles.

**Qualifikationen**:

- Excellent Master’s Degree in engineering, computer science, robotics or any related area
- Excellent programming skills in Python or C++
- Strong knowledge and in-depth understanding of machine learning techniques, especially reinforcement learning, imitation learning, neural networks, and the corresponding software frameworks (e.g., pytorch, tensorflow)
- Experience with Linux and development on Linux Systems
- Fluent English skills, Fluent proficiency in spoken and written German (optional)
- Publication at a machine learning or robotics conference
- Knowledge in the area of robotics and motion planning
- Basic knowledge of ADAS/AD architectures
- Excellent communication skills and desire to work as part of a global team in a multi-cultural environment
- High intrinsic motivation to perform cutting-edge research
- Highly self-organized

**Additional information**:
Would you like to write your doctoral dissertation in cooperation with Mercedes-Benz Group AG? We offer you an international network of experts, research materials, insights into our work and a personal mentor who will serve as your contact partner in addition to your doctoral advisor at your university.