Research Scientist, Embodied AI

The role

You will invent and develop algorithms and systems at the cross-section of computer vision, machine learning, AI, graphics, and robotics. You will advance the Large Geospatial Model and develop new spatial representations for embodied AI.

What you will do

• Develop new spatial representations that embodied AI can learn from, train in, and act on.
• Advance real-to-sim pipelines connecting real-world observations, 3D reconstruction, simulation, and robotic policy learning.
• Solve emerging problems that relate to spatial intelligence, 3D scene representation, training data, model efficiency, and interactive systems.
• Take new capabilities through to evidence.
• Build proof of concept prototypes.
• Collaborate closely with researchers, engineers, and product partners to bring novel models to production.
• Publish and present your work.

Required qualifications

• A PhD in computer vision, robotics, graphics, machine learning, or a related field, or equivalent research experience, with a publication record at top venues.
• Substantial expertise in robotics or embodied AI, including experience in sim-to-real transfer, domain randomization, policy evaluation, and simulation-based learning.
• Deep experience with 3D scene representations: Gaussian splatting, neural fields, meshes, point clouds, or implicit representations.
• Applied generative or diffusion models to 3D data, such as scene completion, inpainting, relighting, or novel view synthesis from sparse capture.
• Worked on 3D semantics, open-vocabulary scene understanding, or scene graphs.
• Proficiency in Python and Deep Learning libraries such as PyTorch.
• GPU or graphics programming experience, such as custom CUDA kernels, shaders, or differentiable rendering.
• A track record of contributing to widely used open-source research codebases.

Nice to have

• Proficiency in C++

Benefits

• Annual bonus
• Equity
• Comprehensive benefits package

About the company

Niantic Spatial builds spatial AI models and services so people and machines can reason, learn and act safely in the physical world. The team builds a living, measured record of the real world that machines can query, learn from and use to understand how places change over time.