FSD Progress Report: Where Tesla's Autonomy Stands Today
An honest assessment of Tesla's Full Self-Driving technology, its capabilities, limitations, and path to true autonomy.
Full Self-Driving is Tesla's most controversial and potentially most valuable technology. After years of development, where does FSD actually stand? Let's cut through the hype and examine the facts.
What FSD Can Do Today
FSD (Supervised) version 12+ represents a major leap forward in Tesla's autonomy capabilities:
Core Capabilities
- City street driving: Navigate complex intersections, traffic lights, stop signs
- Highway driving: Lane changes, merging, exit navigation
- Parking: Parallel and perpendicular, with Summon capabilities
- Unprotected turns: Left turns across traffic (the hardest maneuver)
- Roundabouts: Multi-lane roundabout navigation
- Construction zones: Adapting to temporary road configurations
The v12 Revolution
Version 12 marked a fundamental architecture change. Instead of explicit code for every scenario, Tesla moved to:
- End-to-end neural networks: Video in, steering/acceleration out
- Learned behavior: The car mimics how humans drive
- Fewer hard-coded rules: More natural, human-like driving
This approach has resulted in dramatically smoother driving and fewer "robotic" behaviors.
What FSD Cannot Do (Yet)
Despite the progress, significant limitations remain:
Current Limitations
- Weather: Heavy rain, snow, and fog degrade performance
- Edge cases: Unusual scenarios still confuse the system
- Construction: Complex work zones with human flaggers
- Emergency vehicles: Responding appropriately to police/ambulance
- Confidence: Still requires driver supervision (hands on wheel)
The "Supervised" Reality
Despite the name "Full Self-Driving," Tesla still requires human supervision. The driver must:
- Keep hands on the steering wheel
- Remain attentive and ready to take over
- Take responsibility for the vehicle's actions
Measuring Progress
How do we objectively measure FSD improvement?
Key Metrics
| Metric | Late 2023 | Late 2024 | Improvement |
|---|---|---|---|
| Miles per critical intervention | ~100 | ~500+ | 5x |
| Miles per any intervention | ~20 | ~100+ | 5x |
| User satisfaction | Mixed | Mostly positive | Significant |
These numbers are improving exponentially, roughly doubling every few months.
The Path to Unsupervised
What needs to happen for FSD to become truly autonomous?
Technical Requirements
- Reliability: Interventions must drop to less than 1 per 10,000+ miles
- Safety: Must be statistically safer than human drivers
- Edge case handling: Graceful degradation in unusual situations
- Weather robustness: Reliable operation in adverse conditions
Regulatory Requirements
- Safety data: Compelling evidence of safety vs. human drivers
- Testing: Extensive validation in target jurisdictions
- Approval process: Working with DMVs, NHTSA, international regulators
- Insurance: Frameworks for liability when the car is driving
Tesla's Data Advantage
Why Tesla might win the autonomy race:
Fleet Size
- 6+ million Teslas on the road collecting data
- Billions of miles of real-world driving data
- ~1 million FSD beta users actively training the system
Data Flywheel
The more cars on the road, the more data Tesla collects. More data enables better AI. Better AI sells more cars. This virtuous cycle compounds over time.
Hardware Consistency
Unlike competitors using multiple sensor configurations, Tesla has consistent hardware across its fleet. This simplifies development and deployment.
Competitive Landscape
| Company | Approach | Status |
|---|---|---|
| Waymo | Lidar + HD maps | Operating in limited cities |
| Cruise | Lidar + HD maps | Paused after incidents |
| Tesla | Vision-only, neural nets | Supervised, improving rapidly |
| Mobileye | Camera + radar | Partnerships with OEMs |
| Chinese players | Various | Rapid progress domestically |
Waymo has operational robotaxis today, but Tesla has the scale advantage for eventual dominance.
Timeline Expectations
When might unsupervised FSD arrive? Honest assessment:
- 2025: Continued rapid improvement, possible limited approvals
- 2026: Likely some jurisdictions approve unsupervised operation
- 2027-2028: Broader regulatory approval and robotaxi launch
Note: Tesla has historically been optimistic on FSD timelines. Adjust expectations accordingly.
Investment Implications
For Tesla investors, FSD represents:
- Upside optionality: If solved, worth hundreds of billions
- Current revenue: ~$1-2B annually from FSD purchases/subscriptions
- Competitive moat: Difficult for competitors to replicate data advantage
- Risk: Continued delays erode credibility and investor patience
Conclusion
FSD has made remarkable progress in the past year. Version 12+ demonstrates that end-to-end neural networks can drive in complex environments. But the gap between "impressive demo" and "unsupervised robotaxi" remains significant.
The key question for investors: How much value do you assign to FSD today, given the remaining technical and regulatory hurdles?
Use our simulator to model different FSD adoption scenarios and see how they affect Tesla's potential stock price.
Model Your Own Tesla Scenarios
Use our interactive simulator to test different assumptions and see how they affect Tesla's potential stock price.
Create Your Scenario →