States Are Writing Sensor Checklists Into Self-Driving Laws. The AI Behind the Wheel Is What Actually Matters.

A New Jersey bill would require self-driving vehicles to carry LiDAR and radar alongside cameras, but mandating specific hardware tells regulators nothing about whether the AI actually driving the car is any good.

Across the US, state regulation of autonomous vehicles is aimed at the wrong target. States are racing to pass AV laws, but they’re regulating hardware, not safety. Hardware versus performance is the distinction that will decide whether the next decade of autonomous driving actually gets safer, or just better at satisfying a checklist, and most of the bills moving through statehouses right now get that distinction backward.

The clearest example is a bill from New Jersey, which would require any fully driverless commercial AV to carry cameras plus at least two additional sensing technologies, in practice radar and LiDAR, before it can operate commercially in the state. New York is working on something similar, and other states are watching closely. Under the New Jersey bill, a company could meet every requirement without its AI ever being tested against a real driving benchmark, while a company with a strong safety record could be locked out entirely for choosing a different sensor setup. Neither bill tests which approach, cameras alone or cameras plus LiDAR, actually drives more safely, which is really the only parameter that matters.

The bill regulates what sensors a vehicle carries, never how well it drives, and that’s the flaw. It can’t tell a well-trained model from a poorly trained one as long as both check the same equipment boxes, and it can’t recognize a strong safety record if the vehicle doesn’t carry the required parts. A camera-only fleet that has logged years of incident-free public-road operation could be barred from the state, while a brand-new entrant with a full sensor stack and no track record at all could launch on day one. Meeting the hardware requirement says nothing about whether the AI behind the wheel is any good, or whether that system is safe.

Existing operators face the same problem under the requirement. Some have run autonomous vehicles without a full sensor suite on public roads for years, and the bill would shut them out anyway, discarding years of incident-free driving. What’s more, adding new sensors adds latency, and the more inputs the AI has to reconcile, the longer it takes to arrive at a decision, and a vehicle moving through traffic doesn’t always have that time to spare. Latency is the ultimate constraint of autonomous driving.

Sensors Are Inputs. The AI Decides.

More sensors don’t automatically make a vehicle safer, and fewer sensors don’t automatically make it riskier. What determines how a vehicle drives is the neural networks processing every input in real time. Safety is determined in the AI, not the sensors

LiDAR systems feed the AI pre-measured distance data. Camera-only systems make the AI figure out distance, motion, and intent itself, directly from the visual input. The same way humans do. Either way, the sensors just feed information in, and the AI is what decides how the vehicle actually moves through traffic. Each year, the same AI needs fewer sensors to understand the road than the year before.

McKinsey now says autonomous driving is increasingly defined by generative AI, not by which sensors sit on the roof, and as models improve, they need less help from hardware over time, not more. A hardware mandate locks today’s technology into law no matter how much smarter the AI behind it becomes.

The scale already makes the case. Autonomous vehicle sales are rapidly increasing, with camera-based Tesla recently reaching 1.5m FSD subscribers. The bottom line is that state legislators should be judging AI performance, not sensor lists. Tesla, Volvo, XPeng, and others have all moved toward cameras-only for the same reason. Self-driving’s hard part is no longer what a car can sense, but what its AI-based system can understand, and state law hasn’t caught up.

Who Pays When a Hardware Mandate Outruns the Safety Case?

Multi-sensor systems cost far more to build and maintain than camera-only ones, mainly because of LiDAR hardware, HD-map upkeep, and cloud connectivity. A hardware mandate shrinks the pool of companies that can compete, and it forces every self-driving vehicle, not just robotaxis, to carry that same expensive hardware just to qualify. 

Justifying that cost is hard given how early this industry still is. Despite being a few years and billions of dollars in, most deployments are still limited to a handful of cities on fixed, tightly controlled routes. The World Economic Forum forecasts that by 2035, large-scale robotaxi fleets will still only run in 40 to 80 cities worldwide, mostly in the U.S. and China. Piling state hardware rules on top of already unsettled federal policy just adds another obstacle before any system gets the chance to prove itself. Every year spent litigating sensor requirements state by state is a year not spent proving whether these vehicles are actually safer than the human drivers they would replace, which is the entire point of testing them in the first place.

A hardware mandate answers the wrong question. The real challenge is scaling safe systems onto more roads, faster, and the mandate blocks exactly that while doing nothing to raise the safety bar.

Let Performance Set the Bar

Regulators should set a bar every AV must clear before hitting public roads, but what they should measure, not whether it should exist, is the real question.

Following the introduction of strong edge computing systems in autonomous vehicles, the industry has already moved past costly, single-purpose hardware and is increasingly putting its money into the AI itself. The regulatory conversation should catch up to where the technology already is.

A performance-based standard would let every architecture, camera-only or multi-sensor, prove itself against the same real-world benchmarks, crash rates, near-misses, and performance across weather and traffic conditions. It would also give the public something a sensor checklist never can, namely concrete evidence, not assumptions, that the vehicle stopped next to them at a red light is actually safe.

Rather than more laws deciding which sensors a car must carry, the industry needs one clear, public bar for how well that car has to perform, and regulators willing to let results, not equipment lists pushed by vendors, decide who clears it.