Roboticist, Physical AI

present of AI; San Francisco (CA); On-site; USD 120,000 - 170,000; Full time

Job summary

Sieve in San Francisco seeks a hands-on roboticist to build and operate robot capture rigs and teleoperation setups, ensuring clean, well-labeled sensor data. You will bridge hardware and software to create repeatable data collection pipelines and contribute to multimodal datasets for frontier models.

Ideal candidates have practical integration skills, strong calibration discipline, and comfort turning rough rigs into reliable data pipelines that fuel model training and evaluation.

Qualifications

• Hands-on robotics experience
• Strong calibration and data quality focus
• Ability to turn rough rigs into repeatable pipelines

Responsibilities

• Build and operate robot capture rigs and teleoperation setups
• Log, label and quality-check sensor data
• Integrate hardware with software pipelines for data collection

Skills

Robotics hardware

Sensor calibration

Data quality

Tools

Teleoperation setups

Job description

A roboticist on a physical AI data team typically needs hands on work building and running robots, capture rigs and teleoperation setups, plus the software to log clean, well labeled sensor data. A strong candidate brings practical integration skills across hardware and code, care with calibration and data quality, and comfort turning a rough rig into a repeatable pipeline. Since the company curates video, audio and physical world data for frontier models, the work leans on making real world capture reliable and usable.

How it fits the AI era

Physical AI data is becoming a bottleneck for frontier models, so the people who can build and operate capture hardware matter more, not less. AI speeds up labeling, review and anomaly spotting, but designing rigs, calibrating sensors and judging whether a dataset is trustworthy stay human. The role sits where real world hardware meets model training needs.

How it builds AI experience

This role builds AI experience through hands on work with multimodal datasets, sensor fusion and teleoperation data that feed foundation models. A candidate can prepare by running their own robot or capture setup end to end and documenting data quality decisions. Familiarity with how video, audio and motion data get used in training helps.

Worth showing

• Robot hardware integration
• Teleoperation setup
• Sensor calibration

This is our own read, written from the employer's posting and the company's Spotlight. The employer's posting is the source of truth for the role.

About Sieve

Multimodal artificial intelligence data laboratory creating curated video, audio, and physical world training environments for frontier foundation models, per the company.