Serverless Simulation and a Robot Army of Isaacs

​In physical AI, fast iteration unlocks innovation: data → train → eval → back to data. Every stage needs different hardware, and it can be hard to predict which stage you'll need next.
In this live session, we’ll walk through how robotics teams:

• ​Turn a handful of teleop demos into thousands of sim episodes

• ​Scale from one GPU to multi-node training, and run sweeps in parallel

• ​Evaluate every checkpoint across hundreds of randomized scenes before a real robot moves

​Then we’ll demo it live by training a quadruped robot with RL in simulation using Isaac Lab on Modal.