Lead 3D Perception & Reconstruction Engineer

Orbifold AI, Inc.; Palo Alto, Northern (CA, KY); Hybrid; USD 150,000 - 230,000; Full time

Job summary

Orbifold AI, Inc. is building the infrastructure layer for Physical AI and seeks a researcher to own the 3D reconstruction stack end to end. You will address 6-DoF object pose, hand and full-body pose, multi-view tracking, and calibration challenges across synchronized viewpoints.

You will work with captured data, build automated metrics, and define verification standards for each modality. This role combines deep research with practical validation in a startup-like environment.

Qualifications

• PhD or equivalent in 3D computer vision with first-author publications.
• Expertise in 6-DoF pose estimation, hand/body pose, multi-view or long-horizon tracking.
• Experience with real capture rigs and failure modes such as occlusion and drift.
• Strong PyTorch skills and ability to reason about label quality and model performance.

Responsibilities

• Own the 3D reconstruction stack end to end, including pose and tracking.
• Push accuracy in occlusion, self-contact, fast motion, and reflective/material challenges.
• Train on golden datasets captured in-house with motion capture and force sensing.
• Build automated critics and metrics to catch calibration drift and label noise.
• Define what verified means for each modality and maintain standards.
• Collaborate with partner researchers who audit labels and defend numbers.

Skills

PyTorch

Ray

NeRF

3D reconstruction

Education

PhD in 3D computer vision

Tools

NeRF

3D sensors

Job description

Orbifold AI, Inc. is building the infrastructure layer for Physical AI and seeks a researcher to own the 3D reconstruction stack end to end. You will address 6-DoF object pose, hand and full-body pose, multi-view tracking, and calibration challenges across synchronized viewpoints.

You will work with captured data, build automated metrics, and define verification standards for each modality. This role combines deep research with practical validation in a startup-like environment.