Measure, validated on three independent physics engines.

Today we completed the third and final deep dive of Measure, the instrument in the Senua AI Robotics Core that holds a physics engine to the answers physics gives.

Three engines, one instrument

A result on one engine proves little: a method can be shaped around the engine it was built on. So Measure took three physics libraries from three separate organisations, none connected to Senua AI, and held each to textbook physics, with every rule published before its data existed. Meta's SuperDex Physics: 12 departures, all corrected. Google DeepMind's MuJoCo: 20 issues, 16 fixed and validated. Newton, the Linux Foundation's engine from Disney Research, Google DeepMind and NVIDIA: 29 issues, 23 resolved. 86 pre-registered studies in all, every record public. The issues still open on Newton could have been resolved too; we stopped once the proof across three engines was made.

Built differently

We do not use models. No LLMs. No SLMs. Of any kind.

No GPU. No data centre. No AI token costs, ever. Senua AI learns cause and effect directly from your own system's data, and runs on low-end CPUs: every play in the three deep dives ran on CPUs. What the three engines taught us now goes into the Senua AI Robotics Core. If you build robots or maintain a simulator, talk to us.

The deep dives: Newton · MuJoCo · SuperDex. The open records: Newton · MuJoCo · SuperDex.