Our Research Field
Environment Understanding and Decision Making
What Motivates Us
The real world, and in particular traffic situations, are populated by several interacting individuals all having their own goals. The dynamics of such an environment are highly complex, unknown, and uncertain. Decision making requires understanding of all possible situations and correct inference how own decisions affect other individuals. Planning and decision making must take into account both model-based prediction and their respective uncertainties.
To solve the challenges associated with environment understanding, we focus on the application of probabilistic graphical models as well as recurrent neural networks. By making inference in probabilistic graphical model-based hierarchical environment models, we provide multi-modal behavior predictions of all agents in any given scene and answer questions, such as how likely certain situations are to occur. In combination with reinforcement learning techniques, we determine the optimal decision policies, taking different scenarios and their respective uncertainties into account.
Future driver assistance systems with autonomous capabilities will take over more and more tasks from drivers. This requires robust decision making in environments that are populated by other agents such as cars or pedestrians. At Bosch, we focus on learning appropriate environment models by constructing probabilistic generative models that can be learned from data. Further, they allow for a hierarchical, compressed environment representation optimized for the decision-making and control tasks at stake.
Herman et al."Inverse Reinforcement Learning with Simultaneous Estimation of Rewards and Dynamics"
- Authors: Michael Herman, Tobias Gindele, Jörg Wagner, Felix Schmitt, and Wolfram Burgard
- Published in AISTATS in 2016
Schmitt et al."Predicting Lane Keeping Behavior of Visually Distracted Drivers Using Inverse Suboptimal Control"
- Authors: Felix Schmitt, Hans-Joachim Bieg, Dietrich Manstetten, Michael Herman, and Rainer Stiefelhagen
- Published in IV in 2016
Schmitt et al."I see what you see: Inferring Sensor and Policy Models of Human Real-World Motor Behavior"
- Authors: Felix Schmitt, Hans-Joachim Bieg, Michael Herman, and Constantin A. Rothkopf
- Part of the public project UR: BAN funded by the Federal Ministry for Economic Affairs and Energy
Related Job Openings
Probabilistic Graphical Modeling ResearcherGermany
- Reference no.: DE00558825
Reinforcement Learning ResearcherGermany
- Reference no.: DE00558830
PhD in Risk-averse Reinforcement LearningGermany
- Reference no.: DE00572136