57 minutes ago
1
A Tesla Model S just drove itself across the entire United States without anyone intervening. Not once. Not to dodge a pothole, not to merge onto a freeway, not to navigate a construction zone. The car handled all 3,081 miles from Los Angeles to New York City on its own.
The run took 58 hours and 22 minutes of driving time, averaging 64 miles per hour across some of the most varied road conditions the American highway system has to offer. The vehicle was running FSD version 14.2.2.3 on a 2024 Model S equipped with HW4 hardware. Charging time added another 10 hours and 11 minutes on top of the driving total.
From 32 interventions to zero
The effort was led by Alex Roy, a name that’s been synonymous with Cannonball Run records for years. Roy established the original electric and autonomous Cannonball records back in 2017, so he’s been at this for a while.
In December 2024, Roy’s team completed a coast-to-coast run covering 2,833 miles in 45 hours. That attempt required 32 human interventions.
Then came a subsequent run that covered 2,869 miles from New York City to Los Angeles in 40.5 hours. That one saw only 5 minutes and 20 seconds of total human intervention, translating to 99.78% autonomy.
And now: zero interventions across 3,081 miles. The jump from 32 interventions to five minutes of intervention to literally none happened over the span of a few months and a handful of FSD software updates.
What the rules actually required
The team adhered to strict rules that prohibited any manual intervention during the run. This wasn’t a situation where someone could tap the brake at a yellow light and call it a technicality. The constraints were designed to test whether FSD could genuinely handle end-to-end autonomous driving across a continent.
Context in the autonomy race
The December 2024 run used FSD v12.5.6 and needed 32 interventions. The latest version, v14.2.2.3, needed zero.
Waymo, Tesla’s primary competitor in the autonomous driving space, operates robotaxis in several US cities but relies on pre-mapped geofenced areas. Tesla’s approach is different: train a neural network on vast amounts of driving data and let it generalize to any road, anywhere.
It’s worth noting what this run doesn’t prove. A single successful coast-to-coast trip, even a flawless one, isn’t the same as statistical safety across millions of miles driven by thousands of users. Highway driving, which makes up the bulk of a cross-country route, is fundamentally easier for autonomous systems than dense urban environments with pedestrians, cyclists, and unprotected left turns.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) · 