Robotics Research Scientist: Pre-Training & Data Quality

Google DeepMind; Greater London; On-site; GBP 120,000 - 170,000; Full time

Job summary

Google DeepMind in London is seeking a Research Scientist to setup large-scale tests, deploy ideas quickly, and manage deadlines while applying the latest theories to develop new robotics products, processes, or technologies.

You will work on robot policy training, imitation learning, and pre-training realities, with multimodal data, data labeling, benchmarking, and scalable data mixtures to train robust robotics foundation models.

Qualifications

• PhD or equivalent practical experience in a technical field.
• Experience with algorithmic architectures and training/inference for generative multimodal robotics models.
• Experience optimizing models and systems (performance tuning, experiments, debugging).

Responsibilities

• Drive data quality, acquisition, and labeling strategies, and oversee mixture optimization for robotics foundation models.
• Coordinate training runs, improve training recipes, and hill-climb model performance against benchmarks.
• Build and maintain software tools to iterate on research ideas and verify data quality through robot policy training.
• Contribute to our wider research agenda, including reinforcement learning, vision-language-action modeling, and simulation, in a fast-paced team.

Skills

Algorithmic architectures

Training & inference techniques

Generative multimodal models

Robotics applications

Education

PhD in a technical field or equivalent practical experience

Tools

Simulators

Robot platforms

Benchmarking tools

Job description

Google DeepMind in London is seeking a Research Scientist to setup large-scale tests, deploy ideas quickly, and manage deadlines while applying the latest theories to develop new robotics products, processes, or technologies.

You will work on robot policy training, imitation learning, and pre-training realities, with multimodal data, data labeling, benchmarking, and scalable data mixtures to train robust robotics foundation models.