What’s wrong with Tesla's Full Self-Driving? Insights from Waymo's Founder

Tesla FSD has become shorthand for the promise and peril of modern self-driving technology. When Waymo founder John Krafcik described Tesla’s Full Self-Driving as having a “bad case of myopia,” he crystallized a debate that matters to automotive buyers, owners, and enthusiasts. This article digs into those Waymo insights, explains the technical criticisms, and lays out practical, actionable steps drivers and service shops can take to reduce risk and maintain vehicle systems.

What John Krafcik Means by “Myopia”

Krafcik’s phrase points to Tesla’s commitment to a vision-only approach: relying primarily on cameras and neural nets rather than a mix of sensors like lidar, radar, high-definition maps, and redundant systems. In plain terms, “myopia” highlights the risk of using a single modality (vision) to interpret a world that’s messy and unpredictable.

Why sensor strategy matters

• Redundancy: Multiple sensor types (camera, lidar, radar) increase the chance of correctly perceiving hazards in poor weather, glare, or crowded scenes.
• Edge cases: Rare or unusual situations — construction zones, atypical vehicles, or obscured traffic signals — are handled better with sensor fusion.
• Localization: High-definition maps and precise localization help vehicles predict the environment beyond what cameras can see at a moment in time.

Technical Criticisms in Detail

1. Limited perception in adverse conditions

   Cameras struggle with heavy rain, snow, and direct sun. While machine learning can compensate for many visual distortions, physical sensors like lidar provide depth information that's not affected by contrast or darkness in the same way. For owners, that means a camera-first system may be less reliable under certain weather or lighting conditions.

2. Lack of hardware redundancy

   Industry best practice for safety-critical systems is redundancy: if one sensor fails, another should continue to provide reliable input. Tesla's vision-first approach reduces hardware redundancy, increasing the consequence of a single point of failure.

3. Edge-case generalization

   Self-driving systems are only as good as the edge cases they’ve encountered in training. Without diverse sensor inputs and carefully curated datasets, machine learning models can misinterpret rare situations — the precisely the scenarios Krafcik calls “chaos” that need strong chaos control strategies.

4. Human factors and driver monitoring

   Tesla has used driver-assist features that require human oversight. Effective driver monitoring (attention detection, hands-on-wheel checks, in-cabin cameras) is essential to prevent misuse. Waymo’s approach emphasizes true autonomy in limited domains, reducing human-in-the-loop dependencies.

5. Regulatory transparency and validation

   Waymo has pursued a conservative path with demonstrable system redundancy and controlled geofenced deployments. Critics argue Tesla’s rapid rollout raises questions about third-party validation, logging, and metrics that regulators use to certify safety.

Chaos control: what it is and why it matters

“Chaos control” refers to the ability of a self-driving system to manage highly unpredictable elements on the road: jaywalkers, erratic drivers, temporary traffic patterns, and debris. Krafcik’s point is that controlling chaos requires broader perception, better prediction models, and systems designed to fail safely. For consumers and fleets, weak chaos control increases the likelihood of incidents when the system encounters the unexpected.

What This Means for the Future of Autonomous Vehicles

The debate over vision-only versus sensor fusion isn’t purely academic — it shapes who wins in the race to deploy autonomous vehicles at scale and how fleets and private owners interact with these systems.

• Fleet vs. consumer timelines: Companies using redundant sensor suites and geofencing (like Waymo) may roll out safer, limited-operation services earlier, while consumer-focused vision-first strategies might arrive faster but with higher variability in performance.
• Cost and accessibility: Lidar and other sensors add cost. Vision-first systems aim to scale cheaply, but that cost saving may trade off safety margins in edge conditions.
• Regulation and standards: Expect regulators to demand more transparency and standardized validation tests as incidents accumulate and public scrutiny grows.
• Maintenance and service ecosystems: More complex sensor suites mean new maintenance needs — shops will need calibration tools and expertise to service autonomous-capable vehicles.

Two plausible futures

1. A hybrid world where high-redundancy autonomous taxis operate in geofenced zones while consumer cars keep incremental driver-assist features.
2. A broader adoption of low-cost vision systems that gradually improve, accompanied by stricter regulations and better human-monitoring tools.

Practical Advice for Automotive Buyers and Owners

If you own an EV or are considering one with active self-driving features, follow these actionable guidelines to protect yourself and get the most from the technology.

• Understand the level of autonomy: Ask whether the system is driver assist (Level 2) or truly autonomous (Level 4/5). Do not assume “Full Self-Driving” means hands-off in all conditions.
• Check driver monitoring systems: Make sure the car enforces attentiveness via cameras, sensors, or steering torque monitoring.
• Review software update policies and subscriptions: If FSD capabilities are tied to subscriptions, evaluate cost vs. value over time and read the fine print on function limits. See our primer on Understanding Tesla's Subscription Model.
• Test in realistic conditions: During a test drive, put the system through varied lighting, traffic, and weather to see how it behaves at the edge cases you commonly encounter.
• Insurance and legal considerations: Ask your insurer how advanced driver-assist features affect premiums and claims handling.

Maintenance How-tos: Keep Your Sensors and Systems Reliable

Even the best software depends on clean, well-calibrated hardware. Here’s a step-by-step maintenance checklist you can use to reduce false positives, missed detections, and system downtime.

Daily and weekly checks

1. Visually inspect cameras and sensor housings for dirt, ice, or damage.
2. Wipe lenses with a soft microfiber and a lens-safe cleaner; avoid harsh chemicals that damage coatings.
3. Confirm cabin camera (if equipped) is unobstructed and the windshield area used for sensors is clean.

Monthly and seasonal tasks

1. Run system self-checks in the vehicle UI and note any warning messages.
2. After winter or heavy road salt exposure, perform a deeper clean — follow our Winter-to-Spring Vehicle Deep Clean Checklist to protect sensors and bodywork.
3. Verify camera alignment and sensor calibration after windshield replacements or collisions; these often require dealer or specialist tools.

When to bring the car to a professional

If you see persistent sensor warnings, experience degraded lane-keeping or emergency braking performance, or if a windshield replacement was done without recalibration, schedule a professional diagnostic. Shops need calibration rigs, specialized software, and high-voltage safety training for EVs.

Protective and aftermarket options

Consider manufacturer-approved protective films and sensor covers and ask your installer about optical-grade materials. For EV owners exploring accessories, our guide to accessories and EV trends can point you to compatible options: Top Custom Accessories for EVs and The Future of EVs.

How Service Shops and Technicians Should Prepare

Technicians will play a central role in the autonomous vehicle ecosystem. Shops should invest in:

• Sensor calibration tools and training for lidar/camera/radar alignment.
• Data logging and diagnostics capabilities so they can read black boxes and event logs.
• Safety protocols for handling high-voltage EV systems and data privacy procedures when accessing vehicle software.

Emerging technologies like robotics and automation in repair workflows will change shop operations — our coverage of how robotics are transforming repair shops is a good primer: How Humanoid Robotics Are Revolutionizing the Automotive Repair Landscape.

Final Verdict: A Balanced View

Waymo insights from John Krafcik serve as a healthy critique: Tesla FSD’s vision-first approach brings innovative strengths (rapid iteration, broad real-world data) but also exposes limitations in redundancy and chaos control. For consumers and fleet operators, the takeaway is pragmatic: understand the system’s limits, perform regular maintenance, insist on proper calibration, and don’t mistake marketing language for guaranteed autonomy.

Autonomous vehicles will arrive in stages. Whether they follow a Waymo-style, high-redundancy path or a Tesla-style, vision-first path will determine the near-term shape of the market. In the meantime, owners and shops can take concrete steps to maintain safety and performance while these technologies mature.