Staff Machine Learning Engineer

San Francisco, CA Full-Time $273,000 - $345,000 per year. 2026-06-30

About ATOMS Careers page

Atoms is building the machines that power the next era of progress. They focus on Physical AI, creating real-world robots for critical industries like food, mining, and transport. Their systems are designed to understand, predict, and control the real world with precision, making complex physical operations more reliable, scalable, and productive. This involves deep integration across hardware, software, AI, operations, manufacturing, and real estate, with a commitment to deploying, operating, and improving machines in real environments until they work at scale. Atoms is composed of roboticists, engineers, operators, and builders who believe in transforming physical systems that shape everyday life.

About the Position

Introduction

Atoms is seeking a visionary Staff Machine Learning Engineer to join its founding team. This role is crucial for bridging the gap between high-level AI research and real-world physical actuation for next-generation autonomous transport platforms. We are actively hiring across three core specialized subcategories: AI Research, Post-Training Optimization, and Data Engineering.

Responsibilities

AI Researcher (World Models & VLA)

Research and develop cutting-edge RL and distillation techniques for trajectory planning.
Integrate emerging research from the broader AI community, identifying and prototyping the most promising solutions.
Design and deploy end-to-end multimodal models that translate real-time visual perception and high-level behavioral goals into physical vehicle actuation.
Develop interactive world models from raw multi-sensor logs, allowing the team to re-simulate events and query what a vehicle would see if it altered its trajectory.
Ensure core autonomous driving models can seamlessly adapt to novel urban environments and edge cases.
Partner with validation and QA teams to run model releases through rigorous simulated scenarios, detecting regressions and identifying systemic performance bottlenecks.

Post-Training & Optimization

Own the post-training lifecycle by distilling, quantizing, and optimizing massive models to run with low latency on vehicle edge hardware.
Profile real-time inference pipelines to identify and eliminate CPU, GPU, and memory bandwidth bottlenecks on the vehicle.
Work with low-level hardware, electrical, and firmware teams to iterate on custom carrier boards, sensor interfaces, and GPUs on edge devices.
Benchmark and deploy models utilizing hardware-accelerated runtimes (e.g., TensorRT, CUDA) to minimize inference times under strict constraints.

Data & Long-Tail Scenarios

Architect automated pipelines to ingest, filter, and identify rare, high-value, and long-tail scenarios out of multi-petabyte multi-sensor datasets.
Target and extract complex structural corner cases from real-world driving logs to continuously feed, challenge, and improve our end-to-end behavior models.
Iterate closely with QA, testing, and simulation teams to transform ambiguous real-world anomalies into concrete data blocks for simulation testing.
Implement programmatic data curation, active learning strategies, and statistical quality metrics to optimize the signal-to-noise ratio of our training pipelines.

Requirements

10+ years of non-internship professional MLE experience.

AI Researcher (World Models & VLA)

Deep expertise in applying AI Transformers to robotics, physical actuation, or spatial-temporal data.
Proven track record designing or training multimodal systems, large-scale VLA models, or generative Diffusion models.
Strong background in Sensor Fusion, combining inputs from Cameras, LiDAR, and Radar.
Fluency in PyTorch or JAX for training large-scale models.
Proficiency in Python and familiarity with C++.

Post-Training & Optimization

Strong background in machine learning engineering with a focus on model optimization, distillation, and deployment.
Hands-on experience optimizing models for edge deployment or custom embedded GPU targets.
Deep understanding of profiling tools and debugging resource constraints across CPU/GPU boundaries.
Experience with modern deep learning frameworks (PyTorch or JAX) and runtime compilation.
Robust programming skills in Python and C++.

Data & Long-Tail Scenarios

Professional experience building data curation pipelines, active learning workflows, or data mining architectures for massive physical datasets.
Strong familiarity with robotics data structures and spatial frameworks, including Birds-Eye-View (BEV) or spatial tokenization.
Experience processing and structuring raw data from Cameras, LiDAR, and Radar.
Expert-level proficiency in Python, data engineering frameworks, and PyTorch/JAX.
Exceptional ability to navigate, structure, and derive signal from highly ambiguous, messy, or undefined real-world data distributions.