No, it's because their world model stability & consistency, and sensor input integration into the model is shit. Of course it's not easy. But nobody forced the

No, it's because their world model stability & consistency, and sensor input integration into the model is shit. Of course it's not easy. But nobody forced them to release this half-baked steaming pile. Full visual won't fix the fundamental problem. At best it decreases the cost and energy requirement (but then they'll likely use that for more processing anyway, to try to squeeze more information out of the remaining sensors).