Robots/Parts/AI model
Agility whole-body control model
Agility's own AI model for keeping Digit on its feet and moving its whole body. It handles walking, balance and placing the arms and body where they are told to go, while carrying things. Agility trained it entirely in a simulator and then put it straight onto the real robot. Agility describes it as an always-on safety layer that other, smarter software gives instructions to.
- 1 of 10
- robots in the database
Agility- made by
- AI model
- brain layer
- Not public
- price
Key numbers
- Made by
- Agility, in-house
- Size
- fewer than one million parameters, a small model by today's standards (parameters are the adjustable numbers a model learns)
- Kind of model
- LSTM, a type of neural network with a short-term memory
- How it learned
- trial and error in Nvidia's Isaac Sim simulator, decades of simulated time in three or four days
- What it does
- walking, balance, and moving the hands and body to set positions, including while carrying heavy objects
- Published
- August 2025
Source: Agility, Training a Whole-Body Control Foundation Model (agilityrobotics.com)
Strengths and trade-offs
Our analysisStrengths
Small enough to run quickly on the robot's own computer.
Learned entirely in simulation, so no real robot was risked while it practised.
Lets other software simply say where the hands should go, without worrying about balance.
Agility says it was shown taking instructions from a Gemini research model, so smarter AI can sit on top of it.
Trade-offs
Closed. Only Agility can inspect or change it.
It moves the body but does not decide what task to do. Something else has to tell it.
Only as good as the situations the simulator covered.
A learned controller is harder to explain than written rules when something goes wrong.
Why robot makers choose it
Our readA warehouse robot that falls over is a danger and a cost, so Agility wants one dependable layer that keeps Digit upright whatever else is going on. Training it in simulation is fast and cheap, and keeping it small means it can react instantly on the robot itself.
This is our reasoning from the public facts. No maker has confirmed it.
Robots using it
This list covers only the robots researched so far. A robot missing here may still use this part.
Numbers and listed prices on this page come from the sources shown. Anything marked as our analysis, our read or our estimate is our own view and has not been confirmed by any company. Last checked 14 Oct 2026. How we verify