Who we are
Moving Atoms builds RobotGym. Teams rebuild a workspace from photographs, put a language model at the controls of a simulated robot and download the evidence of every attempt. The physical world is the slow part of robot development. We move the loop into software and we publish the numbers, including the ones we wish were better.
We are two founders, both Oxford mathematicians, and this is the first hire.
What you will do
- Own parts of the simulation stack: MuJoCo scenes compiled from Blender, the 20 Hz control loop, inverse kinematics and the rigid-lift predicate.
- Extend the harness that puts GPT-6 Astra and Claude Fable at the controls: primitives, budgets, the program sandbox and the evidence archive.
- Build environments with design partners from their photographs and drawings, and turn what goes wrong into fixes.
- Write things down. Every limit on the platform is declared, every claim on the site traces to a recorded run, and you will help keep it that way.
What we look for
You do not need every one of these. Strength in most is enough.
- Hands-on work with a robotics simulator such as MuJoCo, Isaac Sim, Gazebo or PyBullet, and the physics behind it.
- Comfort with Python across the stack, from numerical code to HTTP services. Blender scripting or three.js is a plus.
- Experience driving language models through tools or code, and an opinion about what they should and should not be trusted with.
- A habit of measuring before claiming. You would rather report one trial honestly than ten loosely.
- Based in the United Kingdom, or willing to move here.
How we work
Small team, direct communication, short loops. You will talk to design partners yourself. Most decisions are made in the open with the evidence on the table. Salary and equity are set for a founding hire.
How to apply
Email the founders with a note about what you have built and links to anything that runs. A CV is optional. We read everything and reply to everyone.