A bill advancing through the New Jersey legislature would make Tesla's camera-only robotaxi technology illegal in the state — and federal safety investigators have already confirmed the specific engineering failure the bill was designed to prevent.

Senate Bill S1677, cleared by the Senate Transportation Committee on May 11 and now before the Senate Budget and Appropriations Committee, would require every fully driverless commercial vehicle operating on New Jersey roads to carry cameras plus two additional, distinct sensing technologies capable of detecting obstacles even if cameras fail. In practice, that means lidar and radar. In practice, that means Tesla's Cybercab — and the camera-only Full Self-Driving stack underpinning it — cannot enter New Jersey's driverless commercial market without a fundamental hardware overhaul that Elon Musk has spent years publicly ruling out.

What is new this week is the convergence of that bill with a federal investigation that vindicates its premise. In March 2026, the National Highway Traffic Safety Administration escalated its probe into Tesla's Full Self-Driving software to an Engineering Analysis covering an estimated 3.2 million vehicles. The agency's finding was specific: FSD "failed to detect and/or warn the driver appropriately under degraded visibility conditions such as glare and airborne obscurants." Lidar operates by emitting its own laser pulses and measuring their return — it is not degraded by glare because it is not passively receiving reflected sunlight. Radar uses radio waves and penetrates fog and rain that blind cameras entirely. The bill's sensor requirement and NHTSA's failure finding describe the same gap from opposite directions.

Bill Sponsor Is Physicist, Not a Lobbyist

Senator Andrew Zwicker, the bill's primary sponsor, is a physicist at the Princeton Plasma Physics Laboratory — not a career politician responding to industry pressure. He introduced S1677 after riding in a Waymo vehicle in Phoenix and watching sensor fusion in commercial practice. His framing has been deliberate throughout.

"This is not anti-Tesla. I'm pro-New Jersey safety," he told The Verge. The bill's sensor requirement is intentionally technology-neutral — it does not name lidar or radar by brand or by technology type. It requires "a camera system and two distinct sensing modalities that are capable of detecting and tracking obstacles in the event of failure of the camera system." If a future camera-only system can meet that standard, it would qualify. Current Tesla hardware cannot.

The bill also establishes a three-year pilot program for fully driverless commercial vehicles, overseen by the New Jersey Motor Vehicle Commission and the Department of Transportation. Additional provisions include mandatory completion of 50,000 supervised in-state miles before any vehicle may operate without a human monitor, crash reporting requirements, state authorization before launching any commercial driverless service, cybersecurity specifications, restricted operations (no autonomous commercial service in school zones, active construction zones, or areas with elevated pedestrian collision rates), and provisions favoring retention of a steering wheel and pedals — the last of which creates a compounding problem for the Cybercab, a purpose-built two-seat vehicle with neither.

Why Cameras Fail Where Lidar and Radar Don't

The three primary sensing modalities in autonomous vehicles each perceive the world through fundamentally different physical mechanisms, and each has failure modes the others don't share.

Cameras work by passively receiving reflected light. They excel at reading road signs, recognizing lane markings, interpreting traffic signals, and detecting faces and pedestrians under good lighting conditions. They fail when direct sunlight saturates the lens (glare), when water covers the optical surface (rain), or when particulates scatter the light path before it returns (fog, dust, smoke). A camera-based system navigating into a glare-heavy sunset cannot distinguish the road ahead from the sky. The NHTSA finding against Tesla's FSD describes this failure precisely: the system continued operating without warning the driver when camera visibility was critically impaired.

Lidar works differently. A lidar unit emits its own laser pulses — typically in the near-infrared spectrum — and measures the time-of-flight of those pulses as they return from objects in the environment, building a precise three-dimensional point cloud of the surrounding space. Because the system generates its own illumination, it is not affected by ambient glare. Because the laser wavelengths used are not scattered by the same conditions that scatter visible light, lidar degrades in heavy rain and dense fog but not in the sunlit-glare conditions that blind cameras.

Radar uses radio waves — a different region of the electromagnetic spectrum entirely — to measure the velocity and distance of objects. Radio waves penetrate all weather conditions that affect optical sensing: rain, fog, snow, dust, smoke. Radar cannot read signs or detect lane markings, but it can detect that a vehicle ahead has decelerated sharply regardless of whether the sensor can see what caused the deceleration.

Sensor fusion architecture combines all three. The fusion algorithm continuously cross-references what cameras see, what lidar maps, and what radar detects, and flags when they disagree. When cameras are blinded by glare, lidar and radar continue tracking obstacles independently. No single failure can eliminate all obstacle detection simultaneously. That redundancy is the engineering logic behind the bill's two-modality requirement — and it is the engineering logic behind every major commercial robotaxi deployment other than Tesla's.

Professor Philip Koopman, an autonomous vehicle safety expert in Carnegie Mellon's Electrical and Computer Engineering department with more than 25 years in AV safety research, said the camera-only standard is simply not adequate for New Jersey's conditions. "To run 24/7 across the majority of public roads in New Jersey today, it needs lidar," he told The Verge. "It's pretty clear that today camera-only technology is not up to the challenge." Koopman also identified a structural problem with Tesla's small test fleet: a fleet of 42 driverless robotaxis operating under heightened scrutiny will rarely encounter the rare edge-case road conditions that become statistically inevitable as the fleet scales from dozens to thousands of vehicles. The failure modes that don't appear in 800,000 miles of supervised Austin test data will surface, statistically, by the time a fleet reaches tens of millions of miles — at which point they are no longer rare.

Tesla's Counterargument and Its Limits

Tesla's position is not purely defensive. The company's engineering argument is internally coherent: human drivers navigate the world using two cameras — their eyes — and no lidar, and autonomous systems with sufficiently advanced neural networks should eventually achieve the same. Musk has argued that sensor fusion introduces its own risk, because when lidar and cameras disagree on what they detect, the autonomous system must decide which to trust, and that decision is itself a failure point. "Lidar and radar reduce safety due to sensor contention," he wrote on X last year.

Tesla's official response to S1677 went further: "As written, the legislation imposes restrictions so severely that Tesla's autonomous vehicle technology couldn't legally operate in New Jersey. Rather than prioritizing real safety outcomes and performance, the bill specifically bans Tesla from the New Jersey market."

Tesla's preference is for performance-based safety standards — prove safe outcomes through data, using whatever hardware achieves them — rather than prescriptive hardware mandates. That is a legitimate position in the abstract and one many technology policy researchers share. The problem is that Tesla's safety outcome data is contested by the same federal agency evaluating the bill's premise. NHTSA's Engineering Analysis did not find Tesla's camera-only system marginally adequate; it found a specific category of failure — degraded visibility — in which the system could not reliably detect and warn of obstacles. The "safe outcomes through data" argument depends on the data being complete, transparent, and independently verifiable. NHTSA has noted that Tesla is the only ADS operator that makes regular practice of redacting crash narrative details, and that Tesla's published safety statistics use a non-standard methodology that advocacy organizations and independent researchers have argued overstates the comparison to human drivers.

Tesla mobilized its owner base against the bill, messaging New Jersey's 100,000-plus Tesla owners with warnings that the legislation would ban Tesla technology in the state — a characterization that omitted the bill's explicit restriction to fully driverless commercial fleets. The campaign generated roughly 4,000 protest emails to Zwicker's office in a single day, many from owners who believed their personal driver-assist features were at risk. They are not. S1677 does not touch consumer products like Autopilot or FSD Supervised, which require a licensed human driver to monitor the vehicle at all times.

Waymo Arrives Compliant; Tesla Arrives Blocked

The competitive implications of S1677 are direct. Waymo, which uses cameras, lidar, radar, and audio receivers across a fleet of more than 3,000 fully driverless vehicles operating in more than 10 U.S. metropolitan areas at a rate of 500,000-plus paid trips per week, meets the bill's sensor standard on the day it takes effect. Waymo already successfully lobbied to remove an earlier draft provision that would have required a human occupant during driverless phases, further smoothing its path to NJ authorization.

Tesla's position is the reverse. Its 42 fully driverless robotaxis — operating in Austin, Dallas, and Houston under Texas's self-certification framework, with no state-mandated hardware requirements — are camera-only and would not qualify. The Cybercab, now in production at Gigafactory Texas and already launched in Miami, adds no sensors and removes the steering wheel and pedals, creating a compounding legal problem: it would fail both the sensor test and the provisions favoring traditional controls.

The gap in deployment scale reflects years of different risk tolerances and different engineering bets. Musk promised hundreds of thousands of autonomous Tesla vehicles on public roads by the end of 2026. As of July 2026, Tesla operates approximately 31 active driverless vehicles across its markets, of which only about 14 are running unsupervised at any given time. Waymo, which made no such promise, operates at roughly 100 times that operational scale. The Texas permit data tells the same story with precision: Tesla holds 42 authorizations for autonomous operation in Texas; Waymo holds 577 in the same state.

Does Lidar Make Cameras Obsolete? Not Quite.

Sensor fusion does not eliminate cameras — it complements them. Waymo's system uses cameras for visual recognition tasks that lidar and radar cannot perform: reading traffic signs, identifying lane markings, recognizing pedestrian gestures, interpreting traffic signals. What sensor fusion adds is independence. When any single sensor degrades — glare blinds the camera, heavy rain partially scatters lidar returns, or a radar return is ambiguous — the other modalities continue providing independent confirmation of the obstacle environment. The system can detect a collision hazard without requiring all three sensors to agree.

This is the engineering principle that S1677 encodes in law. The bill does not prescribe sensor fusion specifically — it requires two additional sensing modalities capable of detecting obstacles "in the event of failure of the camera system." A future architecture that achieves that standard without traditional lidar or radar hardware would still qualify. The bill is technology-neutral in that specific sense.

Koopman identified the deeper issue with Tesla's approach from a systems safety perspective: the human visual cortex that drives a car is not just a camera. It integrates peripheral vision, stereo depth, motion prediction, pattern recognition across time, and cognitive understanding of other drivers' behavior — a suite of capabilities that has no direct analog in a camera-only neural network that must learn every novel situation from scratch. "Human brains are fundamentally more powerful than AI because we understand," he told The Verge.

What a Map of State AV Law Reveals

There is no federal autonomous vehicle safety standard in the United States. Congress has been unable to pass one since 2017, when the House unanimously approved the SELF DRIVE Act — a bill that would have barred states from setting AV performance standards — and the Senate's companion bill, AV START, died after Democratic senators argued it did not do enough on safety and liability. The failure of that federal legislation created the state-by-state patchwork that governs autonomous vehicles today.

The consequence is a regulatory geography sorted by tolerance for risk rather than by engineering evidence. Texas allows AV operators to self-certify against no mandatory hardware standard and to deploy commercially at Level 4 with no state-mandated supervised testing mileage. California requires detailed permitting and independent review. Arizona is permissive. New Jersey, if S1677 passes, would become the first state to mandate specific hardware redundancy — a sensor standard rather than a performance standard — as a legal prerequisite for commercial driverless deployment.

New York is considering nearly identical legislation. If both Northeast states enact their respective bills, the greater New York metropolitan area — more than 20 million residents, one of the highest-density, highest-revenue ride-hail markets in the country — would be legally closed to camera-only robotaxi services. For Tesla, that is not just a loss of a state market; it is a loss of the northeastern urban corridor that ride-hail economics make most attractive.

The stakes also point in the other direction. If S1677 passes and survives any legal challenge Tesla pursues, the predictable industry response is a renewed push for federal preemption legislation — exactly what the 2017 SELF DRIVE Act attempted. Tesla would not be fighting solely in Trenton; it would be fighting to prevent the state-patchwork model from becoming a national framework, with Congress as the ultimate battleground. A federal preemption bill that replaced prescriptive state mandates with performance-based federal standards would effectively nullify S1677, NY's parallel bill, and any similar legislation enacted by other states. The fight over New Jersey is simultaneously a fight over who sets national AV safety standards.

What Should Readers Watch For

S1677 has not passed. It cleared the Senate Transportation Committee in May 2026 and sits in the Senate Budget and Appropriations Committee, with a full chamber vote expected later this year. Tesla retains time to lobby for amendments, pursue litigation, or — in theory — change its hardware. Its stated engineering philosophy and the Cybercab's production-ready architecture make hardware change unlikely in the near term.

Whether the bill passes in current form, is amended toward performance-based language, or dies in committee will answer a question the AV industry has been arguing about since 2017: whether safety for passengers sharing roads with autonomous vehicles is better assured by what a vehicle must be built with, or by what it must prove it can do.

The current federal evidence suggests the first standard is not satisfied by camera-only architecture as it currently exists. The bill's passage would write that finding into New Jersey law.

Frequently Asked Questions

Does New Jersey's bill ban Tesla from the state entirely?

No. S1677 applies only to fully driverless commercial vehicles — robotaxis operating without a human monitor in the car. It has no effect on Tesla vehicles sold to consumers, Autopilot, or FSD Supervised, all of which require a licensed driver to remain alert and ready to intervene. Tesla owners in New Jersey will not have their driver-assist features affected in any way. What would be prohibited, if the bill passes in current form, is a fully driverless Tesla commercial ride-hailing service.

Why does Tesla not use lidar, and why do competitors disagree with that choice?

Tesla's position, stated by Musk since at least 2019, is that camera-only neural networks are the correct architecture because human drivers navigate using vision alone, and sufficiently advanced AI should do the same. Adding lidar and radar introduces sensor fusion complexity and, Musk has argued, "sensor contention" — the risk that sensors disagree and the system chooses wrong. Competitors including Waymo, Zoox, and others argue the opposite: that independent sensing modalities reduce the risk of undetected failures, because no single failure mode — glare, fog, rain, optical obstruction — can simultaneously disable multiple physically distinct sensor types. NHTSA's Engineering Analysis of FSD found that camera-only systems failed in glare and airborne obscurant conditions. Lidar and radar do not share that specific failure mode.

What does the federal regulatory picture look like for autonomous vehicles right now?

There is no comprehensive federal autonomous vehicle safety standard. The last serious federal attempt, the AV START bill in the Senate, failed in 2017. States have been legislating independently since. NHTSA has opened multiple investigations into Tesla's FSD — most recently an Engineering Analysis covering 3.2 million vehicles — but those are defect investigations, not proactive safety standards. In April 2026, NHTSA relaxed crash-reporting requirements for Level 2 systems, reducing federal oversight at the same moment that state-level bills like S1677 are attempting to increase it.

If New Jersey and New York both pass sensor mandates, what happens to Tesla's expansion plans?

Tesla's Robotaxi service is currently operational in Austin, Dallas, Houston, and Miami. It is pursuing regulatory permission in Phoenix, Arizona, which has a self-certification framework that does not require sensor redundancy. If NJ and NY both pass sensor mandates, Tesla would be legally barred from the entire greater New York metropolitan area unless it adds lidar and radar to its vehicles. Given Musk's repeated statements that lidar is a "fool's errand" and the Cybercab's production design without those sensors, the more likely path — if those bills pass — is legal challenge, congressional lobbying for federal preemption, or market avoidance of the Northeast corridor. None of those outcomes is free of cost to Tesla's commercial expansion timeline.