2026-06-18 👁️ — views

AV Weather & Climate Constraints — The Physical Ceiling on Driverless Expansion

How rain, fog, snow, and heat shape the geographic ceiling for Tesla FSD and Waymo — and which US cities can go driverless first.

Article 26 in the Physical AI Benchmark Series — Weather as the Physical Ceiling

The prior two articles in this series mapped the city-by-city expansion pipelines for Waymo and Tesla, listing weather as one of six gating criteria alongside regulation, HD mapping, fleet density, insurance, and public acceptance. This article deep-dives the weather dimension in full: which conditions still challenge autonomous vehicles, how Tesla and Waymo handle them differently given their opposing sensor architectures, and what the resulting weather map means for where driverless cars can credibly expand by 2030.

Weather is not a secondary constraint. It is a fundamental physical limitation on sensor performance that no amount of software improvement can fully overcome — rain scatters light, snow covers lane markings, and fog reduces visibility regardless of the algorithm trained on top of the sensor. The geographic distribution of US weather is therefore a direct constraint on the geographic distribution of driverless commercial operations.

Section 1 — Weather Condition Severity Table

The table below maps nine weather conditions across five dimensions: how often they occur in US cities, how they affect the two dominant sensor modalities (cameras and LiDAR), how they degrade road marking visibility, and estimated handling capability for Tesla FSD and Waymo. Handling ratings (Good / Moderate / Poor) are estimates based on publicly disclosed sensor architectures, safety reports, and operational deployment geography. Neither company publishes formal adverse-weather performance data.

Reading the table: The two conditions that create the widest Tesla-versus-Waymo performance gap are dense fog and heavy rain — precisely because Tesla relies solely on cameras while Waymo uses sensor fusion. Extreme heat and glare, by contrast, are conditions both players handle well — which is not a coincidence: both companies have prioritized Sunbelt deployments where these are the dominant adverse conditions and heavy snow is absent.

Section 2 — Tesla vs. Waymo: How Sensor Architecture Determines Weather Performance

The sensor stack choice is not a preference — it is a fundamental architectural decision that cascades into every weather performance outcome. Tesla and Waymo have made opposite bets, and each bet carries specific weather tradeoffs.

Tesla (camera-only):

Tesla’s Full Self-Driving system relies exclusively on cameras plus neural network processing to reconstruct a 3D model of the world from 2D image data. This approach, sometimes called “vision-only,” eliminates LiDAR hardware costs but removes a significant layer of sensor redundancy.

In clear conditions, cameras provide rich color, texture, and contextual information that LiDAR cannot match — they can read signs, distinguish traffic light colors, and identify vehicle types at a glance. In adverse weather, however, cameras degrade in predictable ways:

• Heavy rain produces water droplets on lenses that scatter light and blur the image. Windshield wipers help but do not eliminate the problem for all eight camera positions on a Model 3/Y.
• Dense fog limits the effective vision range to below safe stopping distances at highway speeds.
• Camera-only systems have no fallback sensor modality: if the camera input degrades below threshold, there is no LiDAR or radar to maintain situational awareness.

Tesla’s climate advantage is that cameras work well in the sunny, dry markets Tesla has prioritized — Texas, Arizona, California inland. These markets cover a large share of US ride-hailing demand and allow Tesla to accumulate driverless miles at scale before needing to solve the harder weather problems.

Waymo (LiDAR + camera + radar fusion):

Waymo’s sensor stack combines multiple LiDAR units, cameras, and radar, with software that fuses all three modalities into a unified perception model. This is more expensive but creates layered redundancy:

• LiDAR emits its own laser pulses and detects returning reflections — it is less affected by ambient light conditions than cameras. In moderate rain and light fog, LiDAR maintains range while cameras degrade.
• Radar detects objects at range through heavy rain and fog by detecting reflected radio waves. Resolution is lower than LiDAR but the range and weather penetration are superior.
• If cameras degrade significantly, LiDAR and radar continue to provide situational awareness. Waymo’s system can maintain safe operation across a wider band of weather conditions before reaching the “all sensors degraded” threshold.

Waymo’s weather ceiling is not unlimited. LiDAR performance degrades in heavy snow — ice particles scatter the laser beam in ways that water droplets do not. Radar, while weather-penetrating, lacks the resolution to substitute for LiDAR in complex close-range maneuvers. Dense snow remains a genuine challenge for Waymo, not just Tesla.

The key architectural finding: Waymo’s sensor fusion buys approximately one or two weather-severity steps of additional operational range compared to Tesla’s camera-only system. In rain, fog, and light snow, Waymo handles conditions that would challenge Tesla. In heavy snow and ice, both players face unsolved problems.

Section 3 — US City Weather Readiness Map

The table below applies the sensor-architecture analysis to twelve major US cities, estimating current AV readiness and expansion timelines for Tesla and Waymo. Readiness reflects the combination of weather frequency, severity, and match to each company’s current operational capability. All timelines are estimates.

Key observations:

Phoenix and Austin are straightforward: high heat and occasional thunderstorms are within the operational envelope of both players, and both are already live. San Francisco is the most interesting current case — Waymo operates commercially through Bay Area fog because its LiDAR stack degrades more gracefully than cameras, while Tesla’s supervised FSD must rely on software to manage the camera-only fog challenge.

The Chicago-New York-Boston cluster represents the hardest geographical tier: heavy snow, ice, and freezing rain that challenges all current AV sensor architectures. These cities also carry the highest ride-hailing and transit demand of any US markets — the commercial prize is large precisely where the weather problem is hardest.

Section 4 — The Snow Problem: The Last Frontier

Snow and ice are the unsolved problem at the frontier of autonomous vehicle development — for both Tesla and Waymo, and for every AV developer globally. The challenge is not one problem but four compounding problems:

Lane marking erasure. Snow covers painted lane markings — the primary reference cue for both camera and LiDAR guidance systems. When markings are absent, the vehicle must rely on HD map geometry, curb detection, or inference from surrounding traffic. HD maps can be stale; curb detection fails on snow-buried curbs; traffic inference requires other vehicles to follow when there may be none.

Vehicle dynamics unpredictability. AV control systems are tuned on dry and wet pavement. Ice introduces a coefficient of friction that changes minute to minute as temperature and salt concentration vary. A braking maneuver calibrated for wet asphalt can cause a skid on black ice. Solving this requires not just a perception upgrade but a vehicle dynamics model that operates across a wide and continuously measured friction range.

Dynamic snowplow interference. Snowplows are large, slow, irregularly moving vehicles that displace snow in unpredictable patterns and may block multiple lanes simultaneously. They are typically not in HD maps as static objects, and their behavior in mixed traffic is complex. An AV that cannot correctly model a snowplow’s lane-blocking behavior during a heavy snowstorm is not safe in northern cities in winter.

Hardware degradation. Road salt, used extensively in northern US cities from November through March, causes long-term corrosion damage to sensors, particularly LiDAR units mounted low on the vehicle. Extended operation in salted-road environments requires hardware hardening that adds cost and maintenance complexity.

Who is working on snow-capable AV systems:

• Waymo has conducted small-scale winter testing in Michigan. The testing has not advanced to commercial deployment scale, and Waymo has not publicly committed to a timeline for northern city commercial operations.
• Tesla has the largest corpus of snowy-road data via its consumer fleet — millions of vehicles have driven in snow globally, accumulating training data at a scale no dedicated AV program can match. However, this data is from supervised FSD, not driverless operation, and the dynamics problem remains unsolved at the software level.
• Mobileye is developing snow-capable AV systems for European markets, where winter conditions are a commercial requirement rather than an edge case.
• Aurora (the FedEx/Uber Freight AV trucking program) is testing in northern states as part of its commercial freight deployment. Trucking AV faces the same snow challenges on interstates but with less lane-marking dependency than urban environments.

The commercial consensus across all players is that snow-belt northern US cities — Chicago, New York, Boston, Minneapolis, Cleveland — are realistically a 2028–2032 problem for driverless commercial operations, not a near-term one.

Section 5 — Weather as Competitive Geography

The concentration of Tesla and Waymo operations in Sunbelt markets is not an accident. It is a rational, deliberate strategy: operate where the weather ceiling is highest while software matures for harder conditions. The weather map creates a natural expansion sequence that both companies are following, whether explicitly or implicitly.

Phase 1 — Sunbelt first (now):

Texas, Arizona, California inland, and Florida (outside hurricane season) are all within the operational envelope of current AV systems. Heat and glare are manageable; rain is intermittent; snow is effectively absent. This is where both Tesla and Waymo are concentrating commercial deployment in 2026. The addressable market is substantial: these states represent a large fraction of total US ride-hailing demand.

Phase 2 — Mild coastal (2026–2027):

Atlanta, Seattle, and Oregon present manageable rain and fog challenges. Waymo’s sensor fusion gives it an earlier window in persistently rainy Seattle than Tesla’s camera-only system. Neither city presents a snow problem at commercial scale. Atlanta’s thunderstorm season is intense but brief; the baseline operating window is large.

Phase 3 — Variable mid-country (2027–2029):

Colorado, Minnesota dry-cold (below-freezing but low snowfall in some zones), and Texas panhandle (where ice storms occur but are rare) represent edge-case territory. Occasional snow and ice require software robustness but not year-round hardware hardening. Commercial driverless operations in these markets require validated edge-case handling, not a full redesign.

Phase 4 — Snow belt (2030 and beyond):

Illinois, New York, Massachusetts, Michigan, and Ohio represent the hardest geography for AV weather performance. All four compounding snow problems — marking erasure, dynamics unpredictability, snowplow complexity, hardware corrosion — apply simultaneously for months each year. Commercial driverless operations in these markets require a generation of software and hardware development beyond what exists in 2026.

The competitive implication: Tesla’s camera-only architecture creates a marginally narrower near-term weather window than Waymo’s sensor fusion. But the more important constraint for both players is the snow belt — where neither current architecture has validated commercial capability. The snow frontier is an industry problem, not a Tesla-versus-Waymo differentiator, and solving it is the work of the next decade.

Section 6 — About This Series

This is article 26 in the Physical AI Benchmark Series. Previous articles have covered the ramp index, the humanoid race, regulation, capital, compute, sensors, unit economics, the global race, HD mapping, fleet operations, software and OTA, insurance and liability, consumer demand, partnerships, competitive moats, Cybercab versus Model Y, safety data, Waymo Gen 6, Optimus manufacturing, three scorecard snapshots, the 2030 Bear/Base/Bull forecast, the investor framework synthesis, Waymo’s city-by-city expansion pipeline (article 24), and Tesla’s state regulatory map (article 25). This article provides the physical weather dimension that underpins all geographic expansion modeling.

The central finding: weather is not a solvable software problem in the short term — it is a sensor physics constraint that determines which geographies are commercially viable for driverless operations today, which are viable by 2028, and which remain unsolved until the early 2030s. Sunbelt markets are commercially viable now for both players. The snow belt is the industry’s last geographic frontier, and no player has solved it at commercial scale.

Sources

• Waymo safety in adverse weather — Waymo safety report ↗
• Tesla FSD weather performance — Tesla vehicle safety report ↗
• LiDAR performance in adverse weather — IEEE research ↗
• AV snow and ice challenges — RAND Corporation AV research ↗

Tags