00:03 - The feet of the pedestrian are visible

00:04 - The entire body of the pedestrian is visible

00:05 - Collision occurs

If we assume that the computer only had this video as input, then the computer only had 2 seconds to avoid the collision. That would be unavoidable for the computer. But the fact that there was no sign of slowing down or braking in these 2 seconds is pretty alarming.

But it sounds unlikely that this video was the only input to the computer. Did the car have multiple cameras to "see" bright as well as dark objects at night? I would imagine that a self-driving car driving at night would use multiple cameras (tuned to various level of brightness) so that the car can match human level vision?