Institute for Physical AI @ JBI · a demonstration

What decides how to park?

The rule from the last lesson says the rover must drive a curve, but something has to choose that curve, turning a target pose into steering and throttle a hundred times a second. Here that something is a neural network, a real one, driving the rover into the gold parking spot on the board right now. Watch it park. Then guess how big the brain that does this actually is.

Provenance. The policy is a 3→24→24→2 multilayer perceptron, behaviour-cloned from the ρ–α–β pose-regulation controller: the goal in the rover's body frame (ahead, left, heading-error) in, throttle and turn-rate out, run closed-loop on the real nonholonomic model. The constraint stays exact underneath, the wheel still cannot go sideways, so the net must learn the curve. Released open. The demonstration, a policy you can watch park the real kinematics on-device, and count the cost of, is ours.