As a research engineer in the semantic understanding and reasoning group (CR/AIR4) at Bosch Corporate Research, you will develop next-generation methods for training large behavior models for intelligent cyber-physical systems. Your work will focus on how large-scale AI models can acquire robust, generalizable, and goal-directed behaviors through reinforcement learning, multimodal experience, and interaction with learned or simulated environments.
A central part of the role is the use of world models as a foundation for training and validating these systems. In this context, you will investigate how predictive models of environment dynamics, latent state, and agent-environment interaction can support policy learning, planning, behavior synthesis, and evaluation. This includes leveraging world-model-based rollouts for scalable training, using imagined trajectories for efficient policy improvement, and developing validation frameworks that assess generalization, robustness, and safety before real-world deployment.
Your work will bridge foundational research and practical implementation, and will contribute to the design of architectures that connect representation learning, latent dynamics modeling, reinforcement learning, and large-scale behavior modeling. Building the infrastructure needed for pretraining, simulation-based learning, fine-tuning, and benchmarking in Bosch-relevant environments is also part of this role.
The application space spans a broad range of Bosch domains, including robotics, industrial automation, automated driving, and intelligent building or energy systems. You will collaborate closely with AI researchers, robotics experts, control engineers, and domain specialists to ensure that the developed methods are scientifically strong and strategically relevant for real-world Bosch systems.
Your contributions will help establish core Bosch capabilities in scalable behavior learning, model-based reinforcement learning, and physically grounded AI systems that can be trained, validated, and adapted efficiently across applications.