Member of Technical Staff, Robotics Research
San Francisco • FullTime
Posted 1h ago
Job Location
San Francisco
Tech Stack
Remote Work Policy
On-site
Employment Type
FullTime
Categories
AI Research Engineer
About the job
Sieve is seeking its first Robotics Researcher to lead the training and evaluation of world-action models (WAMs) and 3D perception systems. This role focuses on understanding how data shapes robot capabilities, connecting data curation and labeling to measurable gains in robot learning. You will manage projects from inception to completion, identifying model shortcomings, determining data needs, setting up experiments, and testing results on unseen tasks and environments. A significant portion of this work involves 3D, transforming raw multimodal and egocentric video into actionable data for robot learning by incorporating geometry, pose, and action. Simulation will be used as a tool for evaluation and augmentation when beneficial, but not as the primary focus.
Responsibilities
- Compare data sources (egocentric human video, teleop, robot rollouts) and variations in quality, mix, embodiment, and supervision to improve policies and world models.
- Train and post-train action-conditioned world models and policies for tasks like object manipulation and responding to camera motion.
- Develop pipelines for 3D data extraction from video, including depth, camera trajectories, pose, point tracks, and scene reconstruction.
- Design evaluations by defining test tasks, environments, and embodiments to clearly report data-driven model improvements.
- Utilize simulators or learned world models for evaluation, data augmentation, and sim-to-real checks.
- Translate research findings into curation filters, annotation specifications, and capture protocols for data pipelines.
- Collaborate with Sieve engineers and partner lab researchers to identify key problems and share findings.
- Publish research findings through papers, technical reports, blog posts, and benchmarks to influence the field of robot learning data.
Requirements
- 4+ years of research or engineering experience in robot learning, embodied AI, or 3D computer vision (PhD, strong publications, or open-source projects are highly valued).
- Proficiency in Python and PyTorch or JAX for writing, debugging, and modifying training code.
- Hands-on experience with 3D vision tasks such as multi-view geometry, depth estimation, pose estimation, tracking, or reconstruction.
- Experience training and evaluating a policy or world model (e.g., VLA, diffusion policy, imitation learning) on unseen data.
- Strong experimental rigor, including setting baselines, running ablations, and distinguishing data-driven gains from architectural or evaluation improvements.
- Comfort working with large, complex datasets and diagnosing data quality issues.
- Excellent communication skills for clearly explaining findings, tradeoffs, and uncertainties.
- Ability to work in-person at the San Francisco HQ.
Benefits
- 401(k) and full health insurance
- Breakfast, lunch, and dinner covered
- Choice of snacks
- Ubers covered home