From a prompt to a fully working robot policy. SimRig is my attempt to make robot training more accessible and easier to iterate on. This is just the beginning. More features and support for…

Not every behavior needs reinforcement learning. For Franka Panda, the agent built a direct inverse-kinematics controller that traces a five-pointed star—and published the running simulation through…

I think deployable robots need something between raw policy outputs and human-level intent. An adaptive execution layer. Cortex is our first version of that idea.