Research Scientist, Robotics Pre-Training and Data Quality, DeepMind

Research Scientist, Robotics Pre-Training and Data Quality, DeepMind Overview

DeepMind, a leading AI research organization under Google, is seeking a dedicated Research Scientist to focus on robotics pre-training and data quality. This role involves managing large-scale data collection, annotation, and optimization strategies to enhance the training of robotics foundation models. The ideal candidate will have a strong background in algorithmic architectures, multimodal data sources, and training techniques tailored for robotics applications such as visual-language actions and manipulator systems.

Key Responsibilities

• Lead initiatives to improve data quality, acquisition, and labeling processes, ensuring the robustness of training datasets for robotics models.
• Develop strategies for data mixture optimization, managing diverse data sources to maximize training effectiveness.
• Oversee and coordinate large-scale training experiments, refining recipes and systematically improving model performance through iterative hill-climbing against benchmarks.
• Create and maintain software tools and pipelines that enable rapid experimentation and validation of research ideas, with a focus on data quality and system performance.
• Contribute to the broader research efforts in reinforcement learning, vision-language-action modeling, world-action models, and simulation, working collaboratively within a dynamic team environment.
• Share research findings through publications and presentations, collaborating with academic and industry partners worldwide.
• Apply practical experience with robot policy training, including imitation learning and pre-training, to develop robust robotic systems.
• Manage large multimodal datasets, optimizing data mixtures for training models that operate across language, vision, and video modalities, including large language models (LLMs), vision-language models (VLMs), and world models.

Qualifications

• PhD in a relevant technical field or equivalent practical experience.
• Hands-on experience with generative multimodal models, especially in robotics contexts such as VLAs and WAMs.
• Proven ability to optimize models through performance tuning, experimentation, and debugging.
• Experience with iterative performance improvement techniques like hill-climbing.
• Practical knowledge of working with simulators and real-world robotic platforms, including sensing and manipulation technologies.
• Familiarity with benchmarking, data labeling, and data mixture strategies for foundation models.
• Strong motivation to translate research into real-world robotic applications, emphasizing robustness and reliability.
• Ability to develop and maintain software tools that support fast research iteration.
• Experience managing large-scale multimodal datasets for training general models across various modalities.

About the Organization

DeepMind is committed to advancing artificial intelligence to address complex global challenges and accelerate innovative product development. The organization prioritizes safety and ethics in all its projects and fosters a collaborative environment for interdisciplinary research. Researchers at DeepMind have opportunities to publish their work, collaborate internationally, and contribute to impactful AI solutions.

Note: The job posting does not specify whether visa sponsorship or work authorization support is provided. The companyâ??s policies on sponsorship are not explicitly mentioned in the listing.

Degree Requirement: Bachelor’s

Visa Sponsorship May be