Expert Robotics AI (f/m/div.)

• As part of our team, you will own the robotics AI architecture and technical hypotheses, selecting world-action and video-action models, VLA policies, embodied reasoning, or classical planning according to task, data, and execution constraints. • You will design heterogeneous systems that connect learned models with Code-as-Policy programs, Graph-as-Policy structures, and reusable skills. You specify state and action representations, interfaces, execution checks, as well as recovery behavior. • Additionally, you will implement and adapt models and agentic robot programs. You investigate predictive dynamics, imitation learning, reinforcement learning, or feedback-driven skill refinement, then justify your choices with baselines and ablations. • You will also define evaluation across simulation and physical rollouts, including held-out tasks, embodiment transfer, intervention rate, latency, recovery, and repeatability. In doing so, you distinguish first-attempt success from results achieved through retries. • Furthermore, you will collaborate closely with the deployment engineer to turn the architecture into a functioning embodiment. Together, you diagnose model, planning, or control failures and revise the design based on measured robot behavior. • Using the Agentic AI Software Factory, you will drive implementation and experiments while validating generated code and experimental conclusions. You maintain reproducible artifacts, versioned skills, and accountable release decisions. • Education: completed Bachelor’s or Master’s degree in Robotics, Computer Science, Machine Learning, Control Engineering or a related field • Experience and Knowledge:   • substantial hands-on contributions to robotics AI systems on physical robots, with defensible architecture decisions and reproducible results • deep expertise in predictive world-/video-action modeling or agentic robot planning and Code-as-Policy, integrating learned and classical components; relevant experience may include GaP, CaP-X, or ASPIRE/ENPIRE-style feedback loops; comparable implementations are equally accepted, and mastery of every named framework is not required • proven expertise in action grounding, temporal consistency, uncertainty, and failure recovery • strong Python and PyTorch/JAX skills in training/adaptation, inference, and experiment design, combined with the ability to inspect C++ and robot-control interfaces • demonstrated own code or model contributions, supported by meaningful baselines and measured hardware outcomes • terminology familiarity or model-API integration alone is insufficient • Personality and Working Practice: you are an architect who writes and debugs code, challenges weak assumptions, and follows ideas through to physical evidence • Enthusiasm: building robot intelligence that combines prediction, reasoning and reliable execution • Languages:  very good spoken and written German and English https://www.bosch-ai.com   www.bosch.com/research As a global company, we are legally required to comply with applicable EU and U.S. (re-)export control and sanctions regulations. This means that for certain, particularly sensitive positions, we may only hire individuals who have undergone an additional legal screening. This position is one such sensitive role. We will therefore ask you for additional information required for this screening as the application process continues. You can find information about the processing of your data in our Privacy Notice for the application process.  Permanent, open-ended position in the engineering organization for Robert Bosch Robotics GmbH. Apply with your CV and two concise technical case studies: your contribution, architecture, implementation, evaluation and failure analysis. Include robot result