Julian Bernhard, M.Sc.


Guerickestr. 25, 80805 Munich, Germany

E-Mail: bernhard@fortiss.org


Curriculum Vitae

I studied Electrical Engineering at the Technische Universität München with focus on machine learning, control theory and signal processing and graduated with the Master of Science in 2014. Afterwards, I worked as a consulting engineer in the field of signal processing. In 2017, I joined fortiss as a staff researcher in the competence field "autonomous systems and sensor systems".

My research focuses on the application of machine learning methods to the field of behavior generation for autonomous vehicles. Specifically, I am interested in combinations of classical approaches, e.g. search-based methods and Deep Reinforcement Learning for interaction-aware behavior generation. Interaction-aware approaches incorporate the possible reactions of other participants' on the own executed action already during planning. Such behavior is desirable, for instance, in narrow merging scenarios, which require a "negotiation" with other participants. However, when modeling the uncertainty about the behavior of other participants, such approaches are often computationally demanding due to an exponentially increasing set of possible maneuvering options.

Thus, I evaluate how a combination of pretrained desirable behavior, obtained with Deep Reinforcement Learning, and an online search-methods, e.g. Monte Carlo Tree Search, could overcome computational constraints. Further, I look into the uncertainties induced by the different levels of abstractions from environment modeling to solution approximation and what effect these uncertainties have on the safety of the potentially executed action. Hereby, we model the risk of actions via computational models from the field of finance. 


Thesis Opportunities

I am always open for master thesis in one of the presented research fields. Don't hesitate to contact me. Please attach a comprehensive CV, a current transcript of records and some meaningful references proving your coding experience (Python, C++) and the knowledge in the relevant research area (Deep Reinforcement Learning, Probabilistic Methods, Search Algorithms).


MA: Combining Monte Carlo Tree Search and Deep Reinforcement Learning for Autonomous Driving Behavior Generation Tobias Throm
MA: Deep Harsanyi-Bellman Ad Hoc Coordination for Autonomous Driving Behavior GenerationChristoph Caprano2019
MA: Handling Occlusions in Urban Scenarios: A Belief State Planner for Autonomous DrivingNils Quetschlich2018
MA: Intention-Aware Planning for Autonomous Driving using Deep Distributional Reinforcement LearningStefan Pollok2018
MA: Deep Reinforcement Learning for Nonholonomic Path Planning of Mobile Robots in Unstructured EnvironmentsRobert Gieselmann2017