Tesla FSD now avoids potholes

Tesla Full Self-Driving can now avoid potholes

A recent video shared by DirtyTesla shows a Tesla using Full Self-Driving (Supervised) version 14.2.2.5 performing a maneuver that resembles human driving, highlighting how the system can pick up "unwritten rules" it was not explicitly taught.

"Look how FSD scoots over to put the pothole under the car and avoid the tires," the creator said while posting dashcam footage. The vehicle detected a pothole and smoothly adjusted its path so both tires avoided it, allowing the hole to pass beneath the chassis.

Look how FSD scoots over to put the pothole under the car and avoid the tires 🤯🤯🤯 pic.twitter.com/J3w3dnzJUP

— Dirty Tesla (@DirtyTesLa) March 5, 2026

FSD’s newfound ability to avoid potholes

This behavior marks another step for FSD’s end-to-end neural network. Instead of relying on hundreds of thousands of lines of hand-written code for specific cases, newer versions learn from millions of video clips of human driving.

Even without targeted training for "pothole avoidance," the system has seen enough examples of drivers steering around road defects to emulate the action. That can mean gentle lateral moves and, when safe, even crossing a yellow line to protect suspension and tires. This type of response has been observed, including FSD crossing over a yellow line to bypass a large pothole on a one-lane road when no traffic was present, which helps the system feel less rigid and more natural.

Why it won’t always work

FSD is likely not specifically trained for niche scenarios like potholes; it learns many of them from large-scale video of how drivers react. As a result, it may avoid a pothole in one location but fail to do so in another if it lacks sufficient training exposure.

A similar pattern appeared with toll booths. Tesla’s VP of AI explained why this works and that the behavior was not explicitly coded into the neural network. After seeing enough examples, the system learned to stop even without a gate, remain paused when a hand is extended out of the window, and then proceed once the hand is withdrawn—mirroring typical driver behavior. The more examples the model sees, the broader the set of scenarios it can handle. Tesla prioritizes core tasks first—such as detecting and stopping at stop signs—and may later focus on details like potholes.

Beyond the pothole: smarter real-world interactions

Pothole avoidance adds to a run of unscripted improvements. FSD has been seen handling toll booths competently, stopping and waiting for occupants to complete payment before moving on. It has also learned to recognize hand gestures, such as from a road worker or a cyclist signaling the car to proceed.

The system is becoming more responsive to driver intent as well. In some situations, using turn signals can nudge the car to take a different route than the planned one. This comes alongside reports of FSD predicting and avoiding high-speed crashes before they occur.

The road to unsupervised autonomy

As Tesla advances toward a version of FSD that does not require human supervision, its machine-learning approach is at times exceeding expectations. Elon Musk recently said users will soon be able to control FSD with Grok, Tesla’s built-in AI assistant, using natural language to specify where or how to park upon arrival.

With each update, FSD is becoming more confident and capable at managing edge cases once considered out of reach. By learning from millions of miles of human driving, the system is developing a grasp of road nuances beyond what traditional programmed approaches typically deliver.