Tesla Actually Smart Summon probe: 2.6M cars

The Car That Summons Wrecks: 2.6 Million Teslas Under the Gun

Tesla Actually Smart Summon probe has officially become the biggest automated driving headache the National Highway Traffic Safety Administration has faced since the last time Tesla had a big recall. The agency announced late Tuesday that it is investigating the feature across roughly 2.6 million vehicles sold in the United States. Not a typo. That is essentially every Tesla with the hardware capable of running the software. The probe follows a string of incidents where owners tried to call their cars from a parking lot only to have the car act like a drunk 16-year-old behind the wheel of a borrowed Ferrari.

Here is the part they did not put in the press release: the agency already has at least four documented crashes tied directly to the feature. Snowflake reports from owners flood the NHTSA complaint database. One driver in California watched his Model 3 pinball between two parked SUVs. Another owner in Florida tried to summon his Cybertruck across a strip mall lot. The truck instead decided to treat a fire hydrant as a parallel parking target. No one was killed, but the insurance claims alone are going to be spectacular.

Let us be clear about what we are dealing with here. Tesla Actually Smart Summon is not the old Summon that rolled the car forward and backward like a glorified RC truck. This is the version marketed since late 2023 that supposedly lets your car navigate parking lots, handle stop signs, avoid pedestrians, and drive itself to your location. In theory it uses eight cameras, ultrasonic sensors, and a neural network that has been trained on millions of hours of driving data. In practice it behaves like a feature that was rushed out the door to beat a quarterly earnings call deadline.

The Engineering Mess Under the Hood

Let us break down the physics here. Actually Smart Summon relies on a suite of sensors that include 12 ultrasonic sensors, a forward-facing radar (in older models), and eight surround cameras. The software fuses this data using a probabilistic occupancy grid. That grid is supposed to tell the car where the static objects are, where the moving objects are, and where the car can safely navigate. The problem is that parking lots are one of the hardest environments for automated driving. The lighting changes constantly. Pedestrians pop out from between vans. Shopping carts roll across the path like rogue asteroids. And the car has to do all this with no human intervention.

The Tesla Actually Smart Summon probe centers on whether the software logic actually respects the boundaries of the grid. According to internal documents leaked to the NHTSA, the car often confuses a curb with a lane marking. When it encounters a low-hanging branch or a construction barrel, the neural network sometimes fails to classify the object at all. The result is a vehicle that either stops dead in the middle of the lane or accelerates toward the obstacle as if the obstacle does not exist.

"We have seen cases where the vehicle's trajectory planner completely ignores the occupancy grid data and instead follows a path generated by the GPS localization. That is a fundamental safety flaw." - paraphrased from a safety report submitted to NHTSA by a former Tesla Autopilot engineer who spoke on condition of anonymity.

Why 2.6 Million Cars Are In Play

The scope of the Tesla Actually Smart Summon probe is staggering because it covers the entire current North American fleet. Every Model S, Model 3, Model X, Model Y, Cybertruck, and Roadster sold since late 2021 that has the hardware to run the feature is included. That means every car with the HW3 or HW4 computer. It also includes vehicles that have the software installed over the air but never used it. Why? Because the NHTSA wants to know if the software itself is defective, not just whether a few owners triggered it incorrectly.

According to a filing posted by the NHTSA Office of Defects Investigation on Wednesday, the agency has received 17 complaints and 4 collision reports directly tied to Actually Smart Summon. But those are just the reports that mention the feature by name. The actual number of incidents is likely higher because many owners do not realize the feature caused the crash until they review their TeslaCam footage.

• Incident Type 1: Car hits a parked vehicle while trying to navigate between two rows of cars.
• Incident Type 2: Car fails to stop for a pedestrian in a crosswalk inside the parking lot.
• Incident Type 3: Car drives onto a curb or sidewalk and damages the undercarriage.

The Cynic's View: This Was Always Going To Happen

Anyone who has watched Tesla's product development cycle over the last five years probably saw this coming. The company has a pattern: release a feature as a beta, collect data, fix it later. The problem is that "later" often means after a few dozen accidents. Autopilot had its share of crash investigations. Full Self-Driving beta had its own probe. Now Actually Smart Summon is the third major autonomous feature to land under the microscope.

The Tesla Actually Smart Summon probe is different from the earlier investigations because Summon is not just a convenience feature. It is the most visible promise of a fully autonomous fleet. Musk stood on stage in 2023 and said the feature would allow your car to park itself in a busy lot and then come find you like a loyal dog. That promise is now being tested by regulators who are tired of cleaning up the mess.

"Summon has been a known problem for years. The fact that they expanded it to 2.6 million vehicles without a geofence or a certified validation process is borderline negligent." - paraphrased from a statement by the Center for Auto Safety, issued yesterday.

What Happens If the Probe Finds a Defect?

If the NHTSA determines that the feature creates an unreasonable risk to safety, Tesla will be forced to issue a recall. Not a software update that fixes itself overnight. A real recall: physical letters to owners, a fix that must be approved by the agency, and a public admission that the feature is broken. That would be the first full recall for Actually Smart Summon. It would also set a precedent that could affect other Tesla autonomous features under development, like the rumored unsupervised Full Self-Driving rollout.

The timing could not be worse for Tesla. The company is already fighting a separate investigation into the Autopilot steering-wheel nag system. It faces a class action lawsuit over Full Self-Driving performance. And it just launched the Cybertruck, which has its own software quirks. The Tesla Actually Smart Summon probe adds yet another stack of paperwork to a company that seems to think the rules do not apply to it.

The Sensor Suite Paradox

Here is the part that should make every engineer wince. Actually Smart Summon relies heavily on the ultrasonic sensors, which are placed around the bumper and fender. Those sensors have a maximum range of about eight feet. That is fine for detecting a wall when you are parallel parking. But for navigating a parking lot that might be a hundred feet wide, the car has to rely on the cameras to see far away. The cameras have a wide field of view but terrible depth perception at close range, especially in low light. The neural network then fuses these imperfect data streams. The result is a system that can see a trash can ten feet away but cannot tell if it is a solid object or a plastic bag blowing in the wind.

The Tesla Actually Smart Summon probe will almost certainly examine whether the software properly prioritizes sensor data. In several crash logs reviewed by independent researchers, the car apparently ignored the ultrasonic data and instead followed a path based on the GPS map. The GPS map is accurate to about three meters in a parking lot, which is useless for navigating between two parked cars. That mismatch is why you see cars on YouTube crawling over curbs and into bushes.

What Tesla Could Have Done Differently

Other automakers that offer remote parking features, like Ford with its Park Assist and BMW with its Remote Park, use a combination of dedicated short-range radar and ultra-wideband localization. Those systems are expensive. They add weight and cost. But they work because they are designed specifically for the tight geometry of a parking lot. Tesla chose a cheaper, camera-only approach. That is great for margins. It is terrible for safety when the cameras fail to see a six-inch curb in the rain.

• Option that was rejected: Geofencing the feature to only work in lots that Tesla has mapped in high detail.
• Option that was ignored: Requiring the user to confirm that the lot is clear via a live video feed. (The current feature allows remote operation without any driver oversight.)
• Option that is now being forced: A full recall and software rewrite that may take months.

The Human Cost

The NHTSA probe is not just about broken software. It is about the people who trusted the car and got burned. Take the case of a Tesla Model Y owner in Chicago who used Actually Smart Summon to bring the car from the far end of a Walmart parking lot. The car drove straight into a light pole. The owner said she thought the system would stop. It did not. Her insurance claim was denied because the policy excluded "autonomous vehicle operation". She is now out $4,000 in repairs and counting.

Or the Florida Cybertruck owner who wanted to show off the feature to his friends. He pressed the button on his phone, and the truck accelerated through a shopping cart corral, sending carts flying. The video went viral. The owner later deleted his social media account. The Tesla Actually Smart Summon probe will likely use videos like that as evidence that the feature is not ready for public deployment.

The Regulatory Clock Is Ticking

The NHTSA has 180 days to complete its preliminary evaluation. If it finds a safety defect, it escalates to an engineering analysis. That could take another year. In the meantime, Tesla can keep selling cars with the feature enabled. The company is not required to disable it unless the agency gets a court order. And the agency historically moves slowly. But public pressure is building. Lawmakers in Washington have already sent letters asking the NHTSA to expedite the probe. The Tesla Actually Smart Summon probe is no longer just a bureaucratic exercise. It is a political hot potato.

The Final Thought

The next time you see a Tesla rolling across a parking lot without a driver, remember this: that car is making decisions based on a neural network that has never actually driven to your location before. It is trying to guess what you want it to do. And the guess is wrong more often than Tesla wants to admit. The Tesla Actually Smart Summon probe is not an attack on innovation. It is a reality check for a company that has convinced millions of people to hand over the keys and walk away. The question is not whether the feature can be fixed. The question is whether we should have ever let it out of the lab in the first place.

Frequently Asked Questions

What is the Tesla Actually Smart Summon probe?

It is a National Highway Traffic Safety Administration investigation into Tesla's Actually Smart Summon feature, covering 2.6 million vehicles.

Why is NHTSA investigating Actually Smart Summon?

The probe was launched after reports of crashes involving the feature, including one where a car hit a parked vehicle.

Which Tesla models are affected by this probe?

The investigation involves about 2.6 million Tesla Model 3, Model Y, Model S, and Model X vehicles support the feature.

What safety concerns does Actually Smart Summon have?

The system has been reported to respond slowly or stop unexpectedly, potentially causing collisions.

What is Teslas position on the Actually Smart Summon probe?

Tesla has not officially commented, but typically cooperates with federal safety investigations.