A rider getting into one of these cars does not audit the sensor stack. What a rider notices is whether the vehicle turns up and whether the trip ends where it was booked to end. That is why the post Srikanth Thirumalai published on Wednesday is written for somebody else: the legislator drafting a statute and the official signing a permit [1][9].
The claim in it most likely to travel is the last one. Improving a Level 2 driver-assist system toward full autonomy is a "false summit," Thirumalai writes, and true L4 maturity comes only from a purpose-built system validated on closed courses and hardened by driving with no human in the car [10]. Read as an argument about datasets rather than sensors, it says that supervised miles are contaminated by the supervisor: every intervention is a failure that never got recorded as one. Waymo's 200 million miles, on its telling, were driven fully autonomously [3]. Tesla's stated path is the one this attacks directly, that enough data and enough miles turn Autopilot and FSD (Supervised) into a Level 4 platform [11].
The industry count matters more here than the rhetoric. The Verge names four operators, and three of them (Waymo, Zoox, Motional) run lidar alongside radar and camera while Tesla does not [4][12]. That is a tally of the companies in one article rather than a market survey, but it indicates which way a legislature drafting from prevailing practice will lean. New Jersey's pending bill would legalise robotaxis only with multiple sensors on board, which according to The Verge would bar Tesla [9].
Waymo's map argument is a time-to-value argument wearing engineering clothes. HD maps jump-start validation so the service can carry riders from its first trip in a new place, and the map is then treated as one more input, a mental memory alongside the sensors [7]. The cost arrives daily after launch, since every city served is a map to keep current, which is what the AI-driven update pipeline is for [8]. Musk's objection, that maps are a really bad idea because roads change and lanes get repainted, is a maintenance objection dressed as a principle [6].
For anyone actually picking a vendor, the useful grid has two axes. One is whether the operating record was accumulated with a human ready to take over or without. The other is whether a single failed input degrades the system or blinds it, which is the whole point of the redundant world view Waymo describes [3][4]. The quadrant worth buying is unsupervised record plus redundant sensing; the quadrant that gets pitched hardest is supervised record plus a promise of graduation [11]. Neither company hands outsiders the data to place itself on that grid, which is why counting sensors, crude as it is, keeps appearing in draft law: it is the one thing checkable by someone who works at neither firm [9].
