Research Scientist, Spatial AI & Perception

About the role

As a Spatial AI Research Scientist on the Atlas VLA Research team, you will build the perception and geometric reasoning systems that give Atlas a grounded 3D understanding of the world. Your work spans the full spectrum from real-time SLAM and state estimation on humanoid hardware to offline reconstruction pipelines that produce the geometric scene structure used to train and condition large VLM/VLA models.

Responsibilities

• Design and implement real-time SLAM and perception-based state estimation for a mobile humanoid or specialized data collection devices operating in unstructured, dynamic environments
• Build offline 3D reconstruction pipelines (multi-view geometry, SfM/MVS, neural reconstruction, depth/pose fusion) that generate geometric scene structure to inform and supervise large VLM/VLA training
• Pioneer research integrating large VLA and VLM models with 3D spatial perception to enable semantic, language-grounded scene reasoning
• Bridge classical geometric methods and learned approaches - knowing when to use optimization-based estimation versus learned representations, and how to combine them
• Write high-quality, maintainable C++ and Python code that fits into a large production codebase

Requirements

• PhD in Robotics, Computer Vision, Machine Learning, Computer Science, or related fields (or equivalent research experience)
• Prior experience building, and deploying SLAM, visual odometry, or 3D reconstruction systems for robots or autonomous vehicles
• Strong background in one or more of the following: Real-time SLAM, visual-inertial odometry, and state estimation, 3D reconstruction (SfM, MVS, multi-view geometry, neural/implicit reconstruction), Probabilistic state estimation and sensor fusion (factor graphs, filtering, optimization on manifolds), Spatial representations, grounding language/vision into 3D geometry, geometric foundation models, Solid foundation in the math underlying geometric perception (Lie groups, nonlinear optimization, multi-view geometry)
• Strong analytical and debugging skills; ability to write reliable, well-structured research code in C++ and Python

Qualifications

• Nice to have: Experience with modern ML frameworks (PyTorch, JAX) and an understanding of how perception outputs feed large-scale model training, Experience building reconstruction or data pipelines that produce training data for large vision or VLA models, Familiarity with VLA / large behavior models and how spatial grounding improves manipulation and long-horizon behavior, Publications in top-tier computer vision, ML, or robotics conferences (e.g., CVPR, ICCV, ECCV, RSS, ICRA, CoRL)

Skills

• Experience with modern ML frameworks (PyTorch, JAX)
• Familiarity with VLA / large behavior models and how spatial grounding improves manipulation and long-horizon behavior
• Publications in top-tier computer vision, ML, or robotics conferences (e.g., CVPR, ICCV, ECCV, RSS, ICRA, CoRL)

Benefits

The base pay range for this position is between $177,000 to $225,000 annually. Base pay will depend on multiple individualized factors including, but not limited to internal equity, job related knowledge, skills and experience. This range represents a good faith estimate of compensation at the time of posting. Boston Dynamics offers a generous Benefits package including medical, dental vision, 401(k), paid time off and a annual bonus structure. Additional details regarding these benefit plans will be provided if an employee receives an offer for employment.