Member of Technical Staff, Robotics Data

Description

The company is at a turning point: frontier labs want general-purpose robots, and good training data is what holds them back. The team builds the base layer under those models, including data collection systems, operations, an exabyte-scale data warehouse, and a software toolchain for partners.

Model quality depends on data quality. This role owns the question of whether the data can be trusted across every modality collected, including teleoperation on real hardware, egocentric capture, and UMI-style grippers. You catch problems first and help change the process so they stop recurring.

Responsibilities

• Sit between raw collection and usable training data by analyzing it across modalities to find quality problems, such as joint velocity anomalies, bad camera poses, gripper encoder errors, and suspect sessions.
• Build automated validation that runs at ingestion and stops bad data before it reaches the warehouse.
• Design and document data formats and schemas for each collection modality so they stay consistent, versioned, and clear to partners and internal researchers.
• Build visualization tools and dashboards that let the wider team inspect data without writing their own scripts.
• Check cross-modal timing: timestamp sync, dropped frames, and clock drift between camera, joint, and gripper streams.
• Set quality metrics and thresholds for each modality and track whether quality is getting better or worse as collection grows.
• Maintain a shared taxonomy of edge cases and failure modes so everyone uses the same terms for data issues.
• Work with collection operators to find root causes, whether they are systemic (calibration, sensor drift) or specific to an operator.

Requirements

• Baseline: a bachelor's or master's degree, or equivalent experience, in robotics, computer science, mechanical engineering, or a related field.
• Baseline: strong Python for data work (numpy, pandas, matplotlib or plotly) and the ability to handle large, messy datasets.
• Baseline: working knowledge of 3D geometry, coordinate frames, and spatial transforms.
• Baseline: a feel for physical systems, enough to spot a wrong trajectory or joint velocity plot by eye.
• Baseline: experience designing or using structured data formats such as protobuf, HDF5, ROS bags, or similar.
• Good-fit signals (not required): hands-on work with robotics data from a research lab, a robotics startup, or a manipulation or locomotion project.
• Good-fit signals (not required): experience with teleoperation, motion capture, or egocentric data collection.
• Good-fit signals (not required): background in signal processing, sensor fusion, or time-series analysis.
• Good-fit signals (not required): experience building internal visualization tools or dashboards for technical teams.
• Good-fit signals (not required): experience with data versioning, lineage tracking, or schema migration in production.
• Good-fit signals (not required): comfort building things from nothing in early-stage settings.
• Good-fit signals (not required): a strong drive toward robotics. The pace is quick and steady, and most of the time goes to hands-on work.

Conditions

The pace is fast and constant, with a large volume of hands-on work.