The Hardware Reality Check: Why Older Teslas Hit an Autonomy Ceiling
For years, Tesla sold vehicles with the promise that full self-driving capability was just around the corner — a software update away. But in a significant reversal that has rippled through the EV community, Tesla admits older hardware cannot support promised autonomous driving features, specifically the Hardware 3 (HW3) computers installed in vehicles from roughly 2019 through early 2023. The company has acknowledged that these earlier neural-network processors lack the compute throughput and sensor-fusion capacity required for its latest FSD (Supervised) neural-net architecture, let alone any future unsupervised autonomous operation.
This admission forces a critical conversation every EV buyer and owner needs to have: What level of autonomy do consumer vehicles actually offer today, how does that compare to true robotaxi services, and what should you expect from the technology you've already paid for?
SAE Automation Levels — Where Everything Actually Sits
| Level | Name | Who Drives? | Examples |
|---|---|---|---|
| Level 0 | No Automation | Human | Basic cruise control |
| Level 1 | Driver Assistance | Human | Adaptive cruise OR lane-keep |
| Level 2 | Partial Automation | Human (must supervise) | Tesla Autopilot, FSD Supervised, GM Super Cruise, Ford BlueCruise |
| Level 3 | Conditional Automation | System (human on standby) | Mercedes Drive Pilot (limited markets) |
| Level 4 | High Automation | System (no driver needed in geofence) | Waymo, Zoox, Cruise (pre-suspension) |
| Level 5 | Full Automation | System (all conditions) | Does not yet exist commercially |
Source: SAE International J3016
The takeaway is unambiguous: Every Tesla on the road today with Autopilot or FSD (Supervised) operates at SAE Level 2. The driver is legally and practically responsible at all times. True driverless robotaxi service — Level 4 — is an entirely different technology stack deployed by entirely different companies in tightly controlled geofences.
What Tesla's Hardware 3 Limitation Actually Means
Tesla's HW3 computer, introduced in March 2019, features dual custom neural-network processing chips that Tesla designed in-house. At launch, Tesla CEO Elon Musk stated the hardware would be capable of full self-driving without future upgrades. Thousands of customers purchased the $6,000–$12,000 FSD package on that basis.
By late 2024 and into 2025, Tesla's FSD neural-network architecture evolved to require significantly higher inference throughput — processing more camera frames per second through larger, more complex vision models. Tesla's Hardware 4 (HW4), introduced in early 2023, features higher-resolution cameras (5 megapixel vs. 1.2 megapixel), a wider field of view, and roughly 2–3x the compute performance of HW3.
The company eventually acknowledged that HW3 vehicles would not receive the latest FSD improvements and would remain on a legacy branch. For owners who paid thousands for the FSD package expecting continuous advancement toward unsupervised autonomy, this represented a material gap between marketing promises and hardware reality.
Hardware Comparison: HW3 vs. HW4
| Specification | Hardware 3 (2019–2023) | Hardware 4 (2023–present) |
|---|---|---|
| Camera Resolution | 1.2 MP | 5 MP |
| Camera Count | 8 | 7–8 (revised placement) |
| Neural Net Processor | Dual custom chips, ~72 TOPS | Dual custom chips, ~300+ TOPS (est.) |
| Field of View | Narrower forward cameras | Wider forward cameras |
| Radar | Included (early), removed (later) | Removed (vision-only) / HD radar on select |
| Latest FSD Support | Legacy branch only | Full current FSD (Supervised) |
Where True Robotaxi Services Actually Operate in 2026
While Tesla's consumer FSD remains Level 2 driver-assist, Level 4 robotaxi services have been operating commercially — with no driver behind the wheel — for years. Here is where you can actually hail one today:
| Service | Live Cities (2026) | Hours | SAE Level | Geofenced? |
|---|---|---|---|---|
| Waymo One | Phoenix, San Francisco, Los Angeles, Austin (expanding) | 24/7 in most markets | Level 4 | Yes — mapped service areas |
| Zoox | Las Vegas, San Francisco (limited) | Limited hours | Level 4 | Yes |
| Cruise | Limited relaunch in Phoenix (post-suspension) | Restricted | Level 4 | Yes |
| Tesla Robotaxi (announced) | Not commercially available | N/A | Claimed Level 4 (unproven) | N/A |
Waymo, a subsidiary of Alphabet, is the clear leader in commercial robotaxi deployment. The company has completed over 10 million paid driverless trips as of early 2026, according to Waymo's published safety data. Its 6th-generation driver system uses a sensor suite that includes lidar, cameras, and radar — a fundamentally different approach from Tesla's camera-only vision stack.
Technology Stack: Camera-Only vs. Multi-Sensor Fusion
How the Two Approaches Differ
Tesla (Camera-Only / "Tesla Vision"): Since removing radar and ultrasonic sensors from production vehicles in 2021–2022, Tesla relies entirely on 7–8 cameras feeding neural networks for perception, prediction, and planning. The advantage is lower hardware cost and a system that mimics human vision. The limitation is that cameras degrade in heavy rain, fog, direct sun glare, and low-light conditions — and cannot directly measure distance the way lidar can.
Waymo / Zoox (Multi-Sensor Fusion): These Level 4 systems combine high-resolution lidar (which provides precise 3D distance mapping regardless of lighting), multiple cameras, and radar. This redundancy allows the vehicle to cross-check sensor inputs and maintain perception even when one modality is degraded. The trade-off is significantly higher per-vehicle hardware cost — Waymo's 6th-gen sensor suite is estimated to add $30,000–$50,000+ per vehicle.
The hardware gap between Tesla's approach and Waymo's is not merely a philosophical disagreement — it's a fundamental engineering constraint. Tesla's camera-only stack, running on HW3 silicon, simply cannot achieve the perception reliability required for unsupervised Level 4 operation across diverse driving environments. This is precisely why Tesla admits older hardware cannot support promised autonomous driving features at the Level 4 threshold.
Robotaxi Safety Data: What the Numbers Actually Show
Waymo Safety Record (Through Early 2026)
- Total paid driverless trips: 10+ million
- Total autonomous miles driven: 50+ million driverless miles
- Fatalities: 0 as of publication date
- At-fault collision rate: ~85% lower than human benchmark in comparable urban environments (per Waymo's published analysis reviewed by NHTSA)
- Minor contact rate: Declining trend across successive software generations
Note: Safety comparisons depend heavily on operational domain. Waymo operates in geofenced, mapped urban areas — not rural highways or unmapped roads.
For context, the NHTSA reports that approximately 1.35 people die per 100 million vehicle miles traveled in the US overall. Waymo's driverless fleet has not recorded a fatality across tens of millions of miles, though the sample size is still far smaller than the national driving population.
Cruise, by contrast, experienced a high-profile incident in San Francisco in October 2023 in which a pedestrian was struck by a human-driven vehicle and then dragged under a Cruise robotaxi. The incident led to Cruise's California DMV permit suspension and a nationwide operational pause. Cruise has since relaunched in limited form under stricter oversight.
Tesla's FSD (Supervised) safety data is harder to evaluate independently. Tesla publishes periodic safety reports claiming lower crash rates with Autopilot/FSD engaged, but these reports have been criticized by the NHTSA and independent researchers for lacking granular context (road type, weather, time of day, engagement conditions). NHTSA has opened multiple investigations into Tesla Autopilot/FSD-involved crashes, particularly incidents involving stationary emergency vehicles.
Robotaxi Pricing: How It Compares to Uber and Lyft
| Service | Avg. Cost Per Mile | Base Fare | 5-Mile Trip Estimate | Surge Pricing? |
|---|---|---|---|---|
| Waymo One | $1.10–$1.50 | $5.00–$7.00 | $10–$14 | Yes (demand-based) |
| UberX | $1.20–$2.00 | $3.00–$5.00 | $9–$15 | Yes |
| Lyft | $1.15–$1.90 | $3.00–$5.00 | $9–$14 | Yes |
| Zoox (where available) | Introductory promos / comparable to rideshare | Varies | $8–$12 | Limited data |
Robotaxi pricing is currently in the same ballpark as Uber and Lyft — and in some markets, Waymo offers promotional pricing that undercuts both. The long-term thesis is that removing the driver (who typically takes 60–75% of the fare) should make robotaxis significantly cheaper than human-driven rideshare. However, current robotaxi fleets carry high per-vehicle hardware costs, remote-operations overhead, and limited fleet density, which keeps prices competitive rather than dramatically lower.
Current Limitations: What Robotaxis Still Cannot Do
Operational Limitations Across All Robotaxi Services
- Geofencing: Every Level 4 robotaxi operates within a pre-mapped service area. Travel outside the geofence is not possible. Waymo's Phoenix service area, for example, covers roughly 225 square miles — large but finite.
- Weather degradation: Heavy rain, fog, and snow reduce sensor effectiveness. Waymo and others will pause service or restrict routing during severe weather events.
- Construction zones: Unexpected lane closures, temporary signage, and shifted road geometry remain edge cases that can trigger remote-assistance interventions.
- Emergency vehicle interactions: While improving, robotaxis sometimes struggle with complex emergency-vehicle scenarios — pulling over appropriately, navigating around stopped fire trucks with non-standard light configurations.
- Unprotected left turns: High-conflict intersections with no dedicated turn signal remain among the most challenging scenarios, occasionally causing extended wait times or conservative routing.
- Pickup/dropoff precision: Passengers must use designated or approved pickup zones in some markets. The vehicle cannot always stop at an exact street address if no safe pull-over exists.
- No highway driving (most services): Waymo has begun limited highway-capable testing, but the vast majority of commercial robotaxi trips are on surface streets at or below 45 mph.
These limitations are not failures — they are the engineering boundaries within which Level 4 systems can operate safely. The geofence-and-validate approach is precisely why Waymo has a strong safety record: the system only operates where it has been proven capable.
The Regulatory Landscape: Who Approves Robotaxi Operations
Robotaxi deployment in the United States is governed by a patchwork of state and local regulations — there is no single federal robotaxi license.
- California (CPUC + DMV): The California Public Utilities Commission issues permits for commercial passenger robotaxi service, while the DMV issues testing and deployment permits for the autonomous vehicles themselves. Both agencies can suspend permits, as the DMV did to Cruise in October 2023.
- Arizona: Among the most permissive states for autonomous vehicle operations. Waymo launched its first commercial service in Phoenix partly because of Arizona's light-touch regulatory environment.
- Texas: Austin has become a key robotaxi market, with Waymo expanding service there under state legislation that allows autonomous vehicle operation with appropriate insurance and safety requirements.
- Federal (NHTSA): The National Highway Traffic Safety Administration does not "approve" robotaxi services but investigates safety defects, issues recalls, and can take enforcement action. NHTSA's Standing General Order requires reporting of crashes involving Level 2+ and Level 3–5 automated systems.
Tesla's proposed robotaxi service — sometimes referred to as the Tesla Robotaxi or Cybercab platform — would need to navigate this same regulatory gauntlet. As of 2026, no Tesla vehicle has received approval for unsupervised Level 4 commercial passenger service in any US jurisdiction.
What Tesla Owners With HW3 Should Do Now
If you own a Tesla with Hardware 3 and purchased the FSD package, here is the practical reality:
- Your vehicle still has Autopilot and basic FSD (Supervised) features. These include adaptive cruise control, lane-centering, automatic lane changes, and traffic-light/stop-sign recognition. They are Level 2 driver-assist features — you must remain fully attentive.
- You will not receive the latest FSD neural-net improvements. Tesla has placed HW3 vehicles on a legacy software branch. As FSD continues to evolve on HW4 and future hardware, the gap will widen.
- Unsupervised autonomy is not coming to your HW3 vehicle. The compute and sensor limitations are fundamental hardware constraints, not software bugs that can be patched.
- Consider the resale-value impact. As the market increasingly understands the HW3/HW4 divide, HW3 vehicles with the FSD package may not command the premium their original buyers expected.
- If you want current-generation FSD, a hardware upgrade or vehicle trade may be necessary. Tesla has offered HW3-to-HW4 retrofits in limited cases, primarily for employees and select customers. Availability and pricing for general consumers have been inconsistent — check with your local Tesla service center for current options.
Frequently Asked Questions
Is Tesla FSD actually self-driving?
No. Tesla Full Self-Driving (Supervised) is an SAE Level 2 driver-assist system. The driver must remain fully engaged, hands on or near the wheel, and eyes on the road at all times. The "Supervised" designation Tesla added to FSD's name reflects this legal and technical reality. Only Level 4 services like Waymo operate without a human driver.
Where can I use a robotaxi service right now?
As of 2026, Waymo One operates commercially in Phoenix, San Francisco, Los Angeles, and Austin. Zoox has limited service in Las Vegas and San Francisco. Cruise has a restricted relaunch in Phoenix. These services are geofenced — you can only use them within the approved service area.
How much does a robotaxi cost vs. Uber or Lyft?
Current robotaxi pricing is comparable to standard rideshare: roughly $1.10–$1.50 per mile plus a base fare of $5–$7. A typical 5-mile trip costs $10–$14, similar to UberX or Lyft. Surge pricing applies during peak demand.
How safe are robotaxis — is there real data?
Waymo has completed over 10 million paid driverless trips with zero fatalities as of early 2026. Their published data shows an at-fault collision rate approximately 85% lower than the human benchmark in comparable urban environments. Cruise experienced a serious incident in 2023 that led to a nationwide suspension. Independent verification of safety claims remains an ongoing area of regulatory focus.
Can Tesla's older HW3 vehicles ever become true robotaxis?
Based on Tesla's own admissions, no. The Hardware 3 computer lacks the processing capacity and sensor inputs required for unsupervised Level 4 operation. Tesla has acknowledged that HW3 vehicles will remain on a legacy FSD branch and will not support the company's latest neural-network architectures.
Why does Tesla use cameras only while Waymo uses lidar?
Tesla's camera-only approach reduces per-vehicle hardware cost and is based on the thesis that neural networks can extract sufficient depth and spatial information from 2D images — similar to human vision. Waymo, Zoox, and most other Level 4 operators use lidar in addition to cameras and radar because lidar provides direct, high-precision 3D distance measurement independent of lighting conditions. The multi-sensor approach adds cost but provides redundancy critical for unsupervised operation.
