How do they even test this on a model ? I mean it's a multimodal i get that but response time are too big or am i missing something ?
It drives step by step, very slowly.
The course looks like it is something that a human could do in 15 seconds, while Astra took 5 minutes.
By making a simulation first so it can run as slowly as it needs to.
A different way to think of this is, consciousness is just a near real time video game with causal influence.