Arithmancy

Embodied AI needs messier data. Physics validated edge cases for robots before they meet us.

[Arithmancy @ArithmancyAI]

We are excited to share that our data pipeline has generated 5,000 episodes of physics-grounded training data, enough to train a world model and run a reinforcement learning policy entirely in imagination before it ever touches a real robot.

[Arithmancy @ArithmancyAI]

@asimovinc open sourced a humanoid. We gave it a kitchen table.

Describe a scene in plain language. Get valid MJCF, the full Asimov rendered in it, and a pick and place episode with physics ground truth and the robot's belief logged at every timestep.

https://t.co/85w05dIpoM

[Arithmancy @ArithmancyAI]

We’ve been cooking something BIG

Almost ready

[Arithmancy @ArithmancyAI]

@JunyaoShi 👀

[Arithmancy @ArithmancyAI]

Simulation is the key

[clankr @clankrmedia]

Agility Robotics co-founder doesn't believe in the teleop data story.

Jonathan Hurst laid it out on The Cube at the Raise Summit in Paris.

🔵 LLMs learned from internet, robots have no equivalent for movement
🔵 Teleoperation alone would take millions people and centuries to scale.

His answer is simulation, not teleop.

Build a digital twin, train with RL, then transfer to the real robot.

He also talked about Digit's future, and how V5 drops the cage.

🔴 Digit V5 ships by the end of this year, safety is the headline feature
🔴 V5 is built to walk into unmodified, as-built human spaces without one

[Arithmancy @ArithmancyAI]

The sim2real gap race is getting interesting

[Jerry Huang @jerryhuang01]

Everyone in the Bay is building a robot company. Almost nobody is talking about the real bottleneck.

After 2 years running the Robotics Center of Silicon Valley, seeing hundreds of frontier labs and stealth startups, one pattern is undeniable:
The blocker isn't model capability. It’s the sim-to-real loop.

1. Simulation ≠ Hardware Reality A VLA trained in Isaac Sim or MuJoCo hits a wall when it encounters real-world actuator limits: thermal throttling, joint backlash, and sensor drift. Even two units of the exact same robot model have completely different dynamics.
2. Teleop data is inherently off-policy Human demonstrations bootstrap your model, but your model isn't human. That distribution gap doesn't disappear with more of the same data. It only closes through real-world rollouts, failure capture, and targeted data collection.
3. "Working" is undefined until deployment Teams optimize for lab success rates, then fail on real-world edge cases: a wrinkled shirt, 4 PM glare, or a pallet 3cm off spec. Without a real-world evaluation harness, you are just guessing.
4. The solution? Treat RL² as your core loop Internally, we call this Reinforcement Learning in Real Life.
Deploy ➔ capture failures ➔ evaluate against real criteria ➔ collect targeted data ➔ redeploy. Continuous learning on real hardware.
5. The Takeaway The winners in Physical AI won't just have the biggest models. They'll be the teams that can close the real-world learning loop the fastest. Closing the sim-to-real gap is a 2-year infrastructure problem. That’s why we built our stack at the Robotics Center.
6. Come hang out If you're hitting these exact walls, let’s compare notes.

We host robotics builders at 90 Welsh St (SF) every Friday. Drop by.

[Arithmancy @ArithmancyAI]

@jerryhuang01 This is exactly what we are working on, same problem - different approaches

[Arithmancy @ArithmancyAI]

Founding team members have tested and validated the data pipeline

Future is looking very promising

[Arithmancy @ArithmancyAI]

@Kaffchad Thank you

[Arithmancy @ArithmancyAI]

The next robotics advantage will not come from the cleanest demo.

It will come from training against the strange, uncomfortable corners of the real world before deployment.

Arithmancy turns the gap between perception and reality into training data.

Our website is now live: https://t.co/KbndRkCSIX
Let us know what you think

[Arithmancy @ArithmancyAI]

@FabiusDefi A diffusion model
Significant amount of noise
Novelty algorithm

[Arithmancy @ArithmancyAI]

@axisrobotics Really like the Save & Rewind idea.

[Arithmancy @ArithmancyAI]

@AngchenXie Robots that can adjust on the fly will always have an edge in the field!🔥

[Arithmancy @ArithmancyAI]

@AntonioSitongLi Congrats! The concept of this robot really deserve its recognition!

[Arithmancy @ArithmancyAI]

@GMOGroup @GMO_AIR The behind-the-scenes work is where the real breakthroughs happen. Keep it up!

[Arithmancy @ArithmancyAI]

@maggi3wang This is a clever direction! Teaching one missing primitive is a lot more scalable than relearning the whole task.

[Arithmancy @ArithmancyAI]

@techniahq I agree, this is a great concept!

[Arithmancy @ArithmancyAI]

Robot believes:

"The path is clear."

Reality:

A toddler left a toy on the floor 10 seconds ago.

Homes are dynamic.

That's exactly what makes them hard for robots.

[Arithmancy @ArithmancyAI]

@CyberRobooo This is the only way! This kind of environment really tests a robot's reliability rather than a controlled environment.

[Arithmancy @ArithmancyAI]

@abhishekunique7 This is a smart use of simulation.