The company builds physical AI, which means real-world robots for industries such as food, mining, and transport, starting with those three.
Its systems aim to understand, predict, and control physical operations so they become more reliable, scalable, and productive.
The work spans hardware, software, AI, operations, manufacturing, and real estate. The team deploys robots into real sites, runs them, learns from the results, and refines them until they work at scale.
The team includes roboticists, engineers, operators, and builders.
This is a research-focused, hands-on individual contributor role in which you help set the company's long-term AI strategy and technical architecture.
You will help define and build the core machine learning architectures and learning methods behind the company's physical AI systems, working on open-ended problems across foundation models, multimodal learning, world models, computer vision, robotics, and embodied intelligence.
The aim is AI that builds richer representations of the physical world, tracks how environments change, reasons about likely outcomes, and acts on that reasoning.
You are not confined to one predefined technical area and can explore new architectures, learning paradigms, and methods that extend the company's physical AI capabilities.
Responsibilities
Research and build new machine learning architectures, algorithms, and training methods for physical AI.
Push forward foundational work in multimodal learning, representation learning, world models, and embodied intelligence.
Design architectures that learn spatial, temporal, semantic, and physical representations from large datasets.
Create methods for learning from video, images, sensor data, simulation, and other multimodal inputs.
Study self-supervised, generative, predictive, and other learning approaches that deepen models' grasp of the physical world.
Explore how foundation models can reason about changing environments, forecast future states, and inform downstream decisions.
Investigate ways to connect perception, learned world representations, planning, and action generation.
Build architectures and training strategies that improve generalization, robustness, scalability, and learning efficiency.
Plan and run rigorous experiments to test new ideas and understand how models behave.
Create evaluation methods and benchmarks for physical-world understanding, reasoning, prediction, and generalization.
Work with large datasets and distributed training infrastructure to develop and assess more capable models.
Turn promising research into prototypes and work with Applied AI and engineering teams on real-world applications.
Track advances in foundation models, multimodal learning, robotics, and physical AI, and spot chances to move the company's research forward.
Lead technically through research direction, architecture reviews, mentoring, experiments, and hands-on building.
Help shape the research culture, technical standards, and long-term direction of the growing AI group.
Requirements
Deep machine learning expertise, with a record of creating new models, architectures, algorithms, or learning methods.
Solid grasp of modern deep learning architectures and core machine learning principles.
Research experience in at least one area such as foundation models, multimodal learning, representation learning, computer vision, world models, robotics, or embodied AI.
Experience designing and running research experiments, testing hypotheses, and turning results into working models or systems.
Solid understanding of large-scale training and how architecture, data, compute, and performance interact.
Experience with at least one of: self-supervised learning, generative modeling, predictive learning, reinforcement learning, or sequential modeling.
Strong math and theory background in machine learning, optimization, statistics, or a related field.
Hands-on skill in Python, PyTorch, or a comparable machine learning framework.
Proven ability to take on unclear research problems alone and develop new technical approaches.
A record of meaningful research or technical work, shown through deployed systems, new architectures, publications, open-source projects, or other proof of impact.
Ability to explain complex research clearly and work across research, engineering, and robotics.
Demonstrated technical leadership and the ability to shape research direction beyond your own projects.
Relevant experience in foundation models for physical AI, world models, multimodal foundation models, self-supervised and representation learning, video understanding and generative video models, spatial and temporal reasoning, predictive and generative modeling, embodied AI and robot learning, vision-language and vision-language-action models, reinforcement learning and learned decision-making, large-scale model pretraining, and autonomous systems and physical-world intelligence.
Conditions
The role is based in the San Francisco office and is onsite five days a week.
The base salary range is $273,000 to $321,000 per year.
Actual pay is set individually and depends on experience, skills, and qualifications.
Under the San Francisco Fair Chance Ordinance, the company considers qualified applicants with arrest and conviction records.
Applications are taken on a rolling basis.
Benefits
Base salary is one part of total rewards, and you may be eligible for equity awards.
Medical, dental, vision, disability, and life insurance.
Flexible Spending Account or Health Savings Account options.
401(k) plan.
Equity.
Sick time, unlimited flexible time off, and paid holidays.
Paid parental leave.
Pre-tax commuter benefit plan.
Work on a meaningful problem with a team committed to doing it together, in an environment built for doing your best work and growing.
Where you’d work
From the office, 4 or more days a week in the office