Spatial AI Research Engineer: SLAM & 3D Perception

Boston Dynamics; Waltham (MA); On-site; GBP 132,000 - 168,000; Full time

Benefits offered by this job

Medical, dental, vision

401(k)

Annual bonus

Job summary

Boston Dynamics in the United States is seeking a SLAM & Spatial AI Research Engineer for the Atlas VLA Research team. You will advance real-time perception and grounded 3D understanding on Atlas, and develop offline reconstruction pipelines from teleop and robot logs.

You will contribute to integrating large VLA/VLM models with 3D spatial perception, while bridging classical geometry with learned representations in a production C++/Python codebase.

Qualifications

• PhD in Robotics, Computer Vision, ML, CS, or related fields (or equivalent).
• Experience building and deploying SLAM, visual odometry, or 3D reconstruction systems for robots.
• Strong background in real-time SLAM, VO, and 3D reconstruction, plus state estimation.
• Solid foundation in the math of geometric perception (Lie groups, multi-view geometry).

Responsibilities

• Design real-time SLAM and perception-based state estimation for a mobile humanoid or specialized devices.
• Build offline 3D reconstruction pipelines (SfM/MVS, neural reconstruction, depth/pose fusion) for geometric scene data.
• Bridge classical geometry and learned methods; write high-quality C++ and Python code in a production codebase.

Skills

Real-time SLAM

Visual-inertial odometry

Multi-view geometry

Probabilistic state estimation

Lie groups

C++

Python

Geometric perception

Education

PhD in Robotics/CS/CV or related field

Equivalent experience accepted

Tools

PyTorch

Python

C++

Job description

Boston Dynamics in the United States is seeking a SLAM & Spatial AI Research Engineer for the Atlas VLA Research team. You will advance real-time perception and grounded 3D understanding on Atlas, and develop offline reconstruction pipelines from teleop and robot logs.

You will contribute to integrating large VLA/VLM models with 3D spatial perception, while bridging classical geometry with learned representations in a production C++/Python codebase.