Learning Objectives
- Understand how Tesla's vision-only approach to autonomous driving differs from lidar-based competitors
- Identify the capabilities and limitations of FSD Supervised versus fully autonomous driving
- Evaluate where Cybercab fits in the emerging robotaxi landscape alongside Waymo and Zoox
What Is Tesla FSD & Cybercab?
Tesla Full Self-Driving (FSD) is an advanced driver-assistance system that uses a vision-only approach — relying entirely on cameras and neural networks rather than lidar or radar sensors. Approximately 2 million Tesla vehicles on the road today have FSD hardware capability, making it the largest fleet of vehicles with autonomous driving features by a significant margin.
FSD operates in Supervised mode, meaning the driver must remain attentive and ready to take over at all times. The system handles highway driving, city streets, traffic lights, stop signs, lane changes, and parking. Tesla's neural network processes input from eight cameras to build a real-time 3D understanding of the environment, replacing the traditional sensor fusion approach used by competitors like Waymo and Cruise.
Cybercab is Tesla's purpose-built robotaxi — a vehicle designed from the ground up with no steering wheel, no pedals, and no mirrors. Unlike FSD on existing Tesla models, Cybercab is intended for fully autonomous operation in a ride-hailing network. Production began in April 2026, and commercial service began in Austin on September 3, 2026, opening first to shareholders and influencers before wider access through the Robotaxi app. Tesla plans to operate its own fleet alongside licensing the technology to third-party operators.
🎯Tip
Try it today: FSD Supervised is available on all Tesla Model 3, Y, S, and X vehicles with the FSD computer — subscribe monthly or purchase outright from the Tesla app
Pricing
- Cancel anytime
- Transferable between Tesla vehicles on same account
- Tied to the vehicle
- Included in resale value
- Austin service live since September 2026; Tesla has not published a fare card
Tesla has positioned FSD as significantly cheaper than competitors' robotaxi services on a per-mile basis, citing the lower hardware cost of a camera-only system versus lidar-equipped vehicles.
Core Capabilities
Vision-Only Autonomous Driving
Tesla's approach eliminates lidar, radar, and ultrasonic sensors — relying entirely on camera vision processed by custom neural networks running on Tesla's proprietary FSD computer. The system uses occupancy networks that predict the 3D space around the vehicle in real time, identifying vehicles, pedestrians, cyclists, lane markings, and obstacles using visual data alone. This approach allows Tesla to scale across its entire fleet without expensive sensor hardware.
End-to-End Neural Network
FSD has transitioned to an end-to-end neural network architecture where a single model takes raw camera inputs and directly outputs driving commands (steering, acceleration, braking). This replaced the previous modular approach that used separate systems for perception, planning, and control. The end-to-end model learns driving behavior from millions of miles of human driving data collected from the Tesla fleet.
The AI5 Chip: Next-Gen FSD Silicon (April 2026)
Tesla's AI5 chip taped out on April 15, 2026 — roughly two years later than originally planned — and will be the silicon behind FSD v15, a version Musk describes as having 10× the parameters of current FSD (a "Large Model" approach). AI5 also powers the next-generation Optimus brain.
AI5 is dual-sourced during initial ramp: Samsung's Taylor, Texas fab (primary) and TSMC Arizona (secondary). Samsung equipment move-in at Taylor began April 24, 2026; volume 2 nm production is targeted for H2 2027, with the first AI5 vehicles expected in MY2027. Musk has explicitly framed the dual-source strategy as "breaking the foundry monopoly."
Long-term, AI5 and its successors shift to Terafab — the Tesla/SpaceX/xAI/Intel chip foundry joint venture announced March 21, 2026 and sited in Grimes County, Texas, with a $16.8 billion first phase confirmed in August 2026. AI6 is targeted for tape-out in December 2026 (Samsung Taylor exclusive), with roughly doubled performance in the same die area. The vertical-integration play ensures FSD is no longer bottlenecked by external foundry allocation.
Cybercab Robotaxi Platform
Cybercab is a two-seat, purpose-built vehicle with a bidirectional design optimized for autonomous ride-hailing. With no driver controls, the entire interior is passenger space. The vehicle uses inductive (wireless) charging, eliminating the need for physical charging ports. Tesla plans to manufacture Cybercab at high volume to drive per-unit costs below $30,000, targeting operating costs under 20 cents per mile.
The Austin Launch and the Federal Audit (September 2026)
Cybercab began carrying paying riders in Austin on September 3, 2026, five months after production started, with shareholders and influencers riding first and wider access following through the Robotaxi app. The same day, the National Highway Traffic Safety Administration opened an audit query covering an estimated 1,000 Cybercabs. The question is not whether the car drives well but how Tesla certified it: federal motor-vehicle safety standards assume a steering wheel, pedals and mirrors, and Tesla self-certified that the standards which presuppose those controls do not apply. NHTSA is now examining the process and technical data behind that decision.
The precedent matters for anyone planning around Cybercab availability. Zoox self-certified a steering-wheel-free robotaxi in 2022, drew the same kind of audit, and did not receive a federal exemption until July 2026, capped at 2,500 vehicles a year. Tesla says the vehicles comply with every applicable standard and that it will expand gradually.
Strengths
- Massive fleet scale: ~2 million vehicles collecting driving data creates a feedback loop no competitor can match
- Low hardware cost: Camera-only approach eliminates thousands of dollars in lidar and radar sensors per vehicle
- Over-the-air updates: FSD improves continuously via software updates — every vehicle gets the latest model
- Integrated manufacturing: Tesla controls the vehicle, chip, software, and charging network end-to-end
- Consumer availability: FSD is the only advanced autonomous system available for purchase by individual consumers today
Limitations & Considerations
- Supervised only: FSD requires an attentive driver at all times — it is not fully autonomous despite the name
- Regulatory exposure: The Austin service runs on Tesla's own certification that safety standards written for cars with manual controls do not apply, and NHTSA opened an audit of that certification the day the service launched; other jurisdictions have not approved it at all
- Vision-only debate: Critics argue that cameras alone cannot match the depth perception and reliability of lidar in adverse conditions (rain, fog, glare)
- Safety scrutiny: NHTSA and other agencies continue to investigate FSD-related incidents, and the "Full Self-Driving" branding has drawn regulatory criticism
- Cybercab is new as a service: Commercial rides began in September 2026 in a single city, so there is no public safety record yet, and the Zoox precedent shows a certification audit can slow expansion for years
Best Use Cases
| Scenario | Why Tesla FSD / Cybercab |
|---|---|
| Daily commuting | FSD Supervised handles highway and city driving, reducing driver fatigue on routine routes |
| Fleet ride-hailing | Cybercab's purpose-built design and low operating cost target the robotaxi market |
| Long-distance highway driving | FSD excels on highways with lane changes, on-ramps, and traffic navigation |
| Data collection at scale | The Tesla fleet generates more real-world driving data than any competitor |
| Cost-sensitive autonomy | Camera-only approach offers the lowest per-vehicle hardware cost for autonomous capability |
When to choose alternatives:
- Fully driverless rides available today across many cities --> Waymo (paid rides in 14 US cities as of September 2026)
- Purpose-built robotaxi with lidar redundancy --> Zoox (bidirectional, multi-sensor; paid rides in Las Vegas under a federal exemption)
- Commercial trucking autonomy --> Aurora Driver (long-haul freight focus)
- Proven safety record with fully autonomous operation --> Waymo (millions of rider-only miles completed)
Getting Started
- Existing Tesla owners: Open the Tesla app, navigate to Upgrades, and subscribe to FSD for $99/month
- New buyers: Add FSD capability ($8,000) when configuring your Tesla at tesla.com
- Enable FSD Supervised from the Autopilot settings menu on the vehicle touchscreen
- Complete the required safety tutorial before your first FSD drive
- Start with familiar routes to build confidence with the system's behavior at intersections and lane changes
- Cybercab rides are available in Austin through the Robotaxi app; watch Tesla's announcements for other cities, which depend on the outcome of the federal audit
⚠️Warning
Safety reminder: FSD Supervised requires the driver to remain attentive at all times with hands ready to take control. It is a driver-assistance system, not a self-driving system. Never rely on FSD without active supervision.
Key Takeaways
- Tesla FSD uses a vision-only approach across ~2 million vehicles, making it the most widely deployed advanced driver-assistance system in the world
- FSD operates in Supervised mode only — the driver must remain attentive despite the system's capabilities
- Cybercab is a purpose-built robotaxi with no steering wheel, in commercial service in Austin since September 2026 and targeting operating costs under 20 cents per mile, while NHTSA audits how Tesla certified it
- AI5 taped out April 15, 2026 (~2 years late) — powers FSD v15 and next-gen Optimus; dual-sourced at Samsung Taylor + TSMC Arizona; transitions long-term to Terafab
- AI6 tape-out targets December 2026; the Terafab joint venture ensures Tesla is no longer bottlenecked by external foundry allocation
- The camera-only versus lidar debate remains unresolved — Tesla bets on software and data scale while competitors invest in sensor redundancy















