Staff Machine Learning Engineer - Action Models

Atoms•San Francisco, CA

2d•$273,000 - $321,000•Onsite

About The Position

Atoms is building Physical AI—real-world robots for industries like food, mining, and transport. This role focuses on developing models that enable intelligent machines to reason about their environment, make decisions, and translate those decisions into actions in the physical world. It sits at the intersection of machine learning, robotics, autonomous systems, planning, and embodied AI. The position involves exploring how foundation models, world models, and learned representations can be connected to action, moving beyond independently engineered components to models capable of learning sophisticated behaviors. The problems are open-ended, the architecture is evolving, and the systems must function outside of a research environment on real machines in complex physical environments. This is a deeply technical individual contributor role with significant influence over the company's research direction and long-term AI architecture.

Requirements

Deep expertise in machine learning with experience developing models for decision-making, robotics, autonomous systems, or embodied intelligence.
Strong understanding of modern deep learning architectures and their application to sequential decision-making and physical systems.
Experience with one or more areas such as reinforcement learning, imitation learning, behavior learning, planning, control, robot learning, or embodied AI.
Experience developing systems that connect learned representations or perception to downstream actions.
Strong understanding of sequential and temporal modeling and the relationship between actions and future states.
Experience training and evaluating models using large-scale real-world, simulated, or synthetic datasets.
Strong understanding of the full ML lifecycle, including data strategy, model architecture, training, evaluation, optimization, and inference.
Experience translating research ideas into functioning machine learning systems.
Strong software engineering fundamentals and the ability to remain deeply hands-on in Python and/or C++.
Demonstrated ability to operate in ambiguous research spaces where the architecture and solution may not yet be known.
A track record of making consequential technical decisions and influencing research or engineering direction beyond an individual project.
Ability to communicate complex research and technical ideas clearly and collaborate across research, engineering, and robotics disciplines.

Responsibilities

Define and help build Atoms' technical architecture for action models and learned decision-making systems.
Develop models that translate learned representations of the physical world into decisions, plans, and actions.
Explore architectures for reasoning, planning, control, and action generation within complex physical environments.
Develop learned policies and action heads capable of operating across real-world robotics and autonomous systems.
Explore approaches that connect perception and world models directly to downstream decision-making and control.
Research and develop techniques across imitation learning, reinforcement learning, behavior learning, and other data-driven approaches to decision-making.
Explore vision-language-action and other multimodal architectures for physical AI.
Develop approaches that allow models to reason across temporal horizons and understand how actions influence future states.
Train and evaluate models using large-scale real-world, simulated, and synthetic data.
Develop methods for learning from demonstrations, human behavior, robot experience, and other sources of supervision.
Make architectural decisions spanning data, model design, training, evaluation, inference, and deployment.
Establish evaluation methodologies for measuring reasoning, planning, action quality, robustness, and generalization.
Partner closely with researchers and engineers working across perception, world models, robotics, autonomy, simulation, and ML infrastructure.
Translate emerging research in embodied intelligence into systems capable of operating reliably on real machines.
Provide technical leadership through research direction, architecture reviews, mentorship, experimentation, and hands-on engineering.
Help establish the technical bar for the growing AI Research organization and participate in identifying and assessing exceptional engineering and research talent.