the bitter lesson assumes compute is cheap and abundant, which is not really true in most robotics contexts. Unless your datacenter is right next to the robot, and you are streaming real time sensors to it.

The Bitter Lesson comes for everything eventually.

"which is not really true in most robotics contexts. Unless your datacenter is right next to the robot"

Most robots do run where there is internet aka a datacenter right next to them. Otherwise sure, a fable class model is not the right way for making a robot walk forward. But to analyze periodically the state of it - maybe.

Also that compute is fast enough to make split second decisions.