Tesla FSD v15 and the Road to Unsupervised Autonomy: What Must Happen Next | Taha Abbasi

The Biggest Question in Automotive Technology

Taha Abbasi addresses the question that dominates every Tesla conversation: when will Full Self-Driving become truly unsupervised? With FSD v14 proving remarkably capable in real-world testing and v15 on the horizon, the technical gap between supervised and unsupervised autonomy is narrowing. But as Taha Abbasi has learned through his own extensive FSD testing, the remaining distance — while small in percentage terms — represents the hardest engineering and regulatory challenges in the entire autonomous driving effort.

The current state of FSD is impressive. Version 14 handles the vast majority of driving scenarios competently, including complex intersections, construction zones, and highway merging. Taha Abbasi’s testing has shown intervention rates dropping to levels that would have seemed impossible just two years ago. But the standard for unsupervised driving is not competent — it is essentially perfect across billions of miles of diverse conditions.

What v15 Needs to Solve

Based on his analysis of FSD’s current limitations, Taha Abbasi identifies several key areas where v15 must demonstrate improvement to approach unsupervised capability. First, edge case handling: the rare but critical scenarios involving unusual road configurations, ambiguous traffic signals, emergency vehicles, and unpredictable pedestrian behavior. These represent less than 1% of driving situations but account for the majority of remaining interventions.

Second, severe weather performance: FSD must operate reliably in heavy rain, snow, fog, and ice — conditions that challenge even human drivers. Tesla’s camera-only approach faces inherent limitations in low-visibility conditions, and v15 will need to demonstrate robust performance across the full spectrum of weather scenarios that vehicles encounter in normal operation.

The Regulatory Landscape

Even if Tesla achieves the technical capability for unsupervised driving, regulatory approval remains a significant hurdle. As Taha Abbasi has covered in his analysis of UN autonomous driving regulations, different jurisdictions have vastly different frameworks for approving autonomous vehicles. The US regulatory landscape is fragmented, with federal guidance and state-level rules creating a patchwork that Tesla must navigate market by market.

Taha Abbasi notes that Waymo’s approach — deploying in geofenced areas with pre-mapped environments and remote oversight — has been easier to regulate than Tesla’s vision of a vehicle that can drive anywhere without limitation. Tesla’s more ambitious approach offers greater consumer value but requires a higher bar of proof for regulators.

The Data Advantage

Tesla’s strongest argument for achieving unsupervised autonomy is its data advantage. With over 3 million vehicles collecting real-world driving data, Tesla has orders of magnitude more training data than any competitor. Each mile driven by each Tesla contributes to the neural network’s understanding of the driving environment. This fleet learning approach means that every edge case encountered by any Tesla anywhere in the world helps improve the system for all vehicles.

Taha Abbasi compares this to the AI training paradigm more broadly: the companies with the most data and the most compute win. Tesla has both — the fleet provides the data, and the Dojo supercomputer provides the training infrastructure. This combination creates a flywheel effect that competitors cannot replicate without a similarly large fleet of vehicles in the real world.

When Will It Happen?

Taha Abbasi is cautious about predicting specific timelines for unsupervised FSD, noting that Elon Musk’s predictions in this area have historically been optimistic. However, the trajectory is clearly positive: each FSD version represents a meaningful improvement over the last, and the rate of improvement appears to be accelerating as the neural network benefits from increasing data scale.

The most likely path to unsupervised driving, in Taha Abbasi’s assessment, is a gradual expansion: first in limited geographic areas with favorable conditions (likely warm-weather cities with well-marked roads), then progressively expanding to more challenging environments. This mirrors Waymo’s approach but at Tesla’s much larger scale. Whether this happens in 2026, 2027, or later depends on both technical progress and regulatory cooperation — but the direction is inevitable.

---

About the Author: Taha Abbasi is a technology executive, CTO, and applied frontier tech builder. Read more on Grokpedia | YouTube: The Brown Cowboy | tahaabbasi.com