Mechanics
Actuators, structures, hands
We design general-purpose robots as one system: actuators, sensing, control and learning, engineered together instead of bolted together. Hardware we can iterate on like software, and software that knows what the hardware can do.
Our approachJoints, drives and sensing designed around what our control stack needs, so every gain in hardware shows up in behaviour.
Real-time control and state estimation that keep working on uneven floors, with imperfect sensors and real payloads.
Policies trained at scale in simulation, then validated on hardware, with the gap measured rather than assumed.
Every deployment feeds data back into the next model, and every model ships back to every robot.
Actuators, structures, hands
Real-time loops, estimation, safety
Simulation, training, evaluation
Telemetry, updates, data loop