Member of Technical Staff - Human Modeling & Embodied Intelligence

Stealth AI Startup; Redwood City (CA); On-site; USD 150,000 - 230,000; Full time

Benefits offered by this job

Equity share

Health benefits

GPU compute access

Job summary

Stealth AI Startup in California invites a researcher to own the human modeling stack, turning raw video into precise 3D hand and body motion for robot policies. You will set technical direction and help build the team around it.

You will collaborate with robotics engineers on capture rigs, calibration, and retargeting human motion to robot hands and arms, leveraging large-scale GPU compute and state-of-the-art research.

Qualifications

• Deep experience in 3D hand or human pose estimation.
• Fluency with MANO/SMPL-X, camera models, and multi-view geometry.
• Strong PyTorch skills and distributed training experience.
• Rigorous validation of data and baselines before GPU hours.
• PhD in CS, robotics, EE, or equivalent research experience.

Responsibilities

• Build and train models for egocentric 3D hand pose (MANO) and hand–object interaction from head-mounted, wrist, and multi-view cameras.
• Recover whole-body human motion (SMPL-X) in world coordinates from egocentric and third-person video.
• Turn research models into a pipeline that processes thousands of hours of video, with automatic QA and failure detection.
• Define accuracy specs and ground truth, and measure which tracking errors actually matter for robot learning.
• Work with robotics engineers on capture rigs, calibration, and retargeting human motion to robot hands and arms.

Skills

3D hand pose estimation

Motion capture

4D human reconstruction

PyTorch

Distributed training

Data validation

Research experience

Education

PhD in CS/Robotics/EE

Equivalent research experience

Tools

MANO/SMPL-X

Camera models

Multi-view geometry

Job description

We're an early-stage embodied AI company (backed by top-tier VCs) teaching robots to use their hands by learning from human. We turn egocentric and multi-view human video into precise 3D hand and body motion and use it to train robot policies. We operate robot arms, dexterous hands, and teleoperation rigs, and we train on large-scale GPU compute.

The role

You’ll own our human modeling stack: the models and pipeline that turn raw human video into accurate 3D hand, hand–object, and whole-body motion. You’ll set the technical direction for this area and help build the team around it.

What you’ll do

• Build and train models for egocentric 3D hand pose (MANO) and hand–object interaction from head-mounted, wrist, and multi-view cameras.
• Recover whole-body human motion (SMPL-X) in world coordinates from egocentric and third-person video.
• Turn research models into a pipeline that processes thousands of hours of video, with automatic QA and failure detection.
• Define accuracy specs and ground truth, and measure which tracking errors actually matter for robot learning.
• Work with robotics engineers on capture rigs, calibration, and retargeting human motion to robot hands and arms.

What we’re looking for

• Deep experience in 3D hand or human pose estimation, motion capture, or 4D human reconstruction, shown through top-venue papers, widely used code, or shipped systems.
• Fluency with MANO / SMPL-X, camera models, and multi-view geometry.
• Strong PyTorch skills and experience training large vision models, including distributed training.
• Rigor with real-world data: you validate data and baselines before spending GPU hours.
• PhD in CS, robotics, EE, or a related field, or equivalent research experience.

Nice to have

• Experience with egocentric devices or datasets (Project Aria, Quest; Ego-Exo4D, HOT3D, ARCTIC).
• Hand–object contact or grasping, robot learning from human video, or SLAM/VIO.

What we offer

• Founding-team scope and equity
• Salary and health benefits
• Real robots and large-scale GPU compute