Job Description

We expect the intern to display independence and maturity as a researcher, using their experience to construct compelling problem statements, engage in rigorous literature reviews and analyses, design and execute experimental plans, and extract salient insights from the experimental results. To be successful, we expect candidates to have experience in dealing with challenging problems in transfer representation learning and robotics, including: (i) learning safe, robust, or generalizable robot state representations; (ii) designing useful regularization objectives, pretext tasks, or auxiliary objectives; (iii) adapting or transferring representations across different domains (e.g., different embodiments, environments, sim-to-real, tasks, etc.); (iv) dealing with the practicalities related to implementing neural policies, e.g., non-convex optimization “tricks” and multi-machine/multi-GPU parallelized training of large models; (v) conducting careful model performance characterization + error analyses, e.g., determining informative ablations and baselines, inspecting and visualizing learned representations, identifying dataset biases; and (vi) leveraging combinations of human data, robot data, and synthetic data for training robot foundation models to acquire various generalization properties (e.g., generalization across tasks, embodiments, objects, environments, etc.).

Finally, the intern will be expected to contribute to the preparation of industrial patents and to work with teammates to publish a high-quality research paper in a major conference venue.

Tasks

• Discuss relevant tasks in Robotics / Embodied AI and quickly agree on research topic
• Perform extensive literature review, to understand the current state-of-the-art methods
• Generate an R&D plan—detailing the relevant research questions, selected AI/ML task(s), intended methodology, experiments, evaluation metrics, and publication venue targets
• Maintain an on-going report of progress towards goals and present related literature and project status to supervisor(s) and/or colleagues on a weekly basis
• Design and implement proposed methodology, according to the above research plan, while remaining aware of any new developments in the field
• Perform extensive evaluation of the proposed approach and generate results (e.g., proofs, quantitative results that compare the main results with baselines and ablations, qualitative analysis that visualizations the behavioral tendencies of the approach)
• Work with supervisor(s) to publish a high-quality research paper in a major conference venue, e.g., NeurIPS, ICML, ICLR, CoRL, RSS, CVPR/ICCV/ECCV, ICRA, IROS, etc.

Qualifications

Required Qualification:

• Currently pursing a Ph.D. or recent post-doctoral researcher in Robotics, Machine Learning, Computer Engineering, or related
• Minimum GPA of 3.0
• Strong background in Robot Learning, Robotics, and/or Embodied AI
• Extensive experience in training Large Behavior Models, such as VLAs and Diffusion Policies
• Extensive experience in leveraging Large Language Models, Vision-Language Models, and/or foundation models that are grounded with other modalities (e.g., audio, haptics, etc.)
• Extensive experience in training neural models on multi-machine or multi-GPU setups
• Extensive experience in working on robot hardware platforms
• Extensive publication history in top conference venues

Desired Qualification:

• Extensive experience in development in Python on Linux-based systems
• Theoretical background in ML topics, e.g., transfer representation learning, non-convex optimization, reinforcement learning, provably safe learning, etc.

Additional Information

Equal Opportunity Employer, including disability / veterans.

BOSCH is a proud supporter of STEM (Science, Technology, Engineering & Mathematics)

• FIRST Robotics (For Inspiration and Recognition of Science and Technology)
• AWIM (A World In Motion)

For more information on our culture and benefits, please visit:

Culture and Benefits | Bosch in the USA