Tesla's Full Self-Driving (FSD) system is crossing a critical threshold. Recently, a Model 3 equipped with FSD v14 embarked from Los Angeles on the U.S. West Coast, traversed the entire continent, and arrived in South Carolina on the East Coast within 2 days and 20 hours. Covering a total distance of 2,732 miles, the journey relied 100% on FSD, navigating complex scenarios including highways, city streets, night driving, and multiple entries and exits from Supercharger stations, all without a single instance of human intervention.
This was not an official demonstration or a laboratory test, but a real-world driving record completed by an ordinary car owner in actual traffic conditions. For the autonomous driving industry, the significance of this journey extends far beyond simply "driving a long distance"; it makes a question realistic and concrete for the first time: Can FSD already completely replace a human driver? Zero-Intervention Cross-Country Trip The Tesla owner who completed this cross-U.S. journey was Davis Moss. According to Moss's posts on social media platform X, his Model 3 was equipped with AI4 hardware, running FSD v14.2.1.25. Data from the FSD database and community trackers indicated that before completing this coast-to-coast drive, Moss had already used FSD for 10,638.8 miles, relying on it 100% throughout. Moss departed from the Tesla restaurant in Los Angeles and ultimately arrived in Myrtle Beach, South Carolina, using FSD v14 for the entire journey without any form of human takeover. This 2,732.4-mile trip covered diverse American road environments, including highways, city streets, and various traffic conditions. Moss emphasized that FSD v14.2 not only completed the entire drive but also handled all parking maneuvers, including automated parking at Tesla Supercharger stations, with "not even a single close call" occurring during the entire trip.
Elon Musk himself promptly shared and congratulated the achievement. It is worth noting that this specific route is the very goal Musk has repeatedly mentioned since the release of Autopilot 2.0 in 2016, but had failed to deliver on until now. Back then, he predicted that Tesla could achieve "coast-to-coast" autonomous driving by 2017. In hindsight, that goal was not impossible.
The Tesla community responded enthusiastically to this achievement, as a zero-intervention coast-to-coast drive has long been considered a significant marker of autonomous driving technology maturity. Tesla's official North American account confirmed on social media: "The first Tesla to drive itself coast to coast using FSD Supervised. Zero interventions. 100% FSD."
Passing the "Physical Turing Test"? Coinciding with this event, Jim Fan, Head of AI Agents at NVIDIA, offered an intriguing judgment: Tesla FSD v14 might have already passed the "Physical Turing Test." This concept originates from the classic Turing Test proposed by mathematician Alan Turing in 1950 but shifts the evaluation criteria from text-based dialogue to physical behavior in the real world. If an observer cannot distinguish whether a task was performed by a human or a machine, the machine is considered to have passed the test. After a test ride in a vehicle using FSD v14, Fan's impression was that it had become very difficult to tell the difference. In his experience, Tesla's driving behavior did not resemble a robot rigidly following rules, but instead acted like a cautious, experienced human driver—creeping forward at intersections, braking smoothly, changing lanes naturally, and responding to subtle cues that are difficult to implement through manual programming. Fan believes that passing the physical Turing test requires overcoming four major challenges: understanding 3D space, finely processing objects, mastering real-world background knowledge, and bridging the gap between digital commands and physical actions. Driving恰好combines all four of these challenges, making it one of the hardest problems in embodied AI to solve. The Advantage of 7 Billion Miles of Data The breakthrough performance of FSD v14 stems from Tesla's shift from a rule-based system to an end-to-end neural network. Early autonomous systems relied on hard-coded rules, whereas FSD v14 is trained on billions of miles of real-world driving data, learning driving patterns in a manner akin to humans. Tesla states that its FSD-equipped vehicles have now accumulated nearly 7 billion miles of driving, with approximately 2.5 billion of those miles completed in urban environments. Urban driving is far more complex than highway driving, involving unprotected turns, unpredictable pedestrians, cyclists, traffic signals, and adverse weather conditions—scenarios where autonomous systems typically face the greatest difficulty. A recent case further demonstrated the system's capabilities: a video released by a Tesla owner showed FSD driving continuously for 7 hours through a severe hailstorm with extremely poor visibility and significant road flooding, without any human intervention. This performance makes the concept of "human-like driving" feel more tangible and real. Fan compares this transition to the proliferation of smartphones: initially astonishing, then becoming the norm, and ultimately indispensable. If machines can move and behave as naturally as humans in the physical world, it would open the door for robots that understand intent rather than merely follow commands. However, Tesla's system still requires human supervision, and even its most ardent supporters acknowledge that perfection is unattainable. Whether driven by a human or a machine, cars inherently involve risk. The system is currently still defined as "FSD Supervised," requiring the driver to remain alert and prepared to take over control at any moment.
Comments