looking under the hood

US wants details on Tesla's fix from recalled automated driving system

The Austin, Texas, company supposedly fixed its self-driving software for more than 2 million vehicles, but the U.S. National Highway Traffic Safety Administration still has concerns. Photo courtesy of Tesla

Federal highway safety investigators want Austin-based Tesla to tell them how and why it developed the fix in a recall of more than 2 million vehicles equipped with the company's Autopilot partially automated driving system.

Investigators with the U.S. National Highway Traffic Safety Administration have concerns about whether the recall remedy worked because Tesla has reported 20 crashes since the remedy was sent out as an online software update in December.

The recall fix also was to address whether Autopilot should be allowed to operate on roads other than limited access highways. The fix for that was increased warnings to the driver on roads with intersections.

But in a letter to Tesla posted on the agency's website Tuesday, investigators wrote that they could not find a difference between warnings to the driver to pay attention before the recall and after the new software was released. The agency said it will evaluate whether driver warnings are adequate, especially when a driver-monitoring camera is covered.

The agency asked for volumes of information about how Tesla developed the fix, and zeroed in on how it used human behavior to test the recall effectiveness.

Phil Koopman, a professor at Carnegie Mellon University who studies automated driving safety, said the letter shows that the recall did little to solve problems with Autopilot and was an attempt to pacify NHTSA, which demanded the recall after more than two years of investigation.

“It’s pretty clear to everyone watching that Tesla tried to do the least possible remedy to see what they could get away with,” Koopman said. “And NHTSA has to respond forcefully or other car companies will start pushing out inadequate remedies.”

Safety advocates have long expressed concern that Autopilot, which can keep a vehicle in its lane and a distance from objects in front of it, was not designed to operate on roads other than limited access highways.

Missy Cummings, a professor of engineering and computing at George Mason University who studies automated vehicles, said NHTSA is responding to criticism from legislators for a perceived lack of action on automated vehicles.

“As clunky as our government is, the feedback loop is working,” Cummings said. “I think the NHTSA leadership is convinced now that this is a problem.”

The 18-page NHTSA letter asks how Tesla used human behavior science in designing Autopilot, and the company's assessment of the importance of evaluating human factors.

It also wants Tesla to identify every job involved in human behavior evaluation and the qualifications of the workers. And it asks Tesla to say whether the positions still exist.

A message was left by The Associated Press early Tuesday seeking comment from Tesla about the letter.

Tesla is in the process of laying off about 10% of its workforce, about 14,000 people, in an effort to cut costs to deal with falling global sales.

Cummings said she suspects that CEO Elon Musk would have laid off anyone with human behavior knowledge, a key skill needed to deploy partially automated systems like Autopilot, which can't drive themselves and require humans to be ready to intervene at all times.

“If you're going to have a technology that depends upon human interaction, you better have someone on your team that knows what they are doing in that space,” she said.

Cummings said her research has shown that once a driving system takes over steering from humans, there is little left for the human brain to do. Many drivers tend to overly rely on the system and check out.

“You can have your head fixed in one position, you can potentially have your eyes on the road, and you can be a million miles away in your head,” she said. “All the driver monitoring technologies in the world are still not going to force you to pay attention.”

In its letter, NHTSA also asks Tesla for information about how the recall remedy addresses driver confusion over whether Autopilot has been turned off if force is put on the steering wheel. Previously, if Autopilot was de-activated, drivers might not notice quickly that they have to take over driving.

The recall added a function that gives a “more pronounced slowdown” to alert drivers when Autopilot has been disengaged. But the recall remedy doesn’t activate the function automatically — drivers have to do it. Investigators asked how many drivers have taken that step.

NHTSA is asking Telsa “What do you mean you have a remedy and it doesn’t actually get turned on?” Koopman said.

The letter, he said, shows NHTSA is looking at whether Tesla did tests to make sure the fixes actually worked. “Looking at the remedy I struggled to believe that there’s a lot of analysis proving that these will improve safety,” Koopman said.

The agency also says Tesla made safety updates after the recall fix was sent out, including an attempt to reduce crashes caused by hydroplaning and to reduce collisions in high speed turn lanes. NHTSA said it will look at why Tesla didn't include the updates in the original recall.

NHTSA could seek further recall remedies, make Tesla limit where Autopilot can work, or even force the company to disable the system until it is fixed, safety experts said.

NHTSA began its Autopilot investigation in 2021, after receiving 11 reports that Teslas using Autopilot struck parked emergency vehicles. In documents explaining why the investigation was ended due to the recall, NHTSA said it ultimately found 467 crashes involving Autopilot resulting in 54 injuries and 14 deaths.

Trending News

A View From HETI

A new generation of technology is making it faster, safer, and more cost-effective to identify CUI. Courtesy photo

Corrosion under insulation (CUI) accounts for roughly 60% of pipeline leaks in the U.S. oil and gas sector. Yet many operators still rely on outdated inspection methods that are slow, risky, and economically unsustainable.

This year, widespread budget cuts and layoffs across the sector are forcing refineries to do more with less. Efficiency is no longer a goal; it’s a mandate. The challenge: how to maintain safety and reliability without overextending resources?

Fortunately, a new generation of technologies is gaining traction in the oil and gas industry, offering operators faster, safer, and more cost-effective ways to identify and mitigate CUI.

Hidden cost of corrosion

Corrosion is a pervasive threat, with CUI posing the greatest risk to refinery operations. Insulation conceals damage until it becomes severe, making detection difficult and ultimately leading to failure. NACE International estimates the annual cost of corrosion in the U.S. at $276 billion.

Compounding the issue is aging infrastructure: roughly half of the nation’s 2.6 million miles of pipeline are over 50 years old. Aging infrastructure increases the urgency and the cost of inspections.

So, the question is: Are we at a breaking point or an inflection point? The answer depends largely on how quickly the industry can move beyond inspection methods that no longer match today's operational or economic realities.

Legacy methods such as insulation stripping, scaffolding, and manual NDT are slow, hazardous, and offer incomplete coverage. With maintenance budgets tightening, these methods are no longer viable.

Why traditional inspection falls short

Without question, what worked 50 years ago no longer works today. Traditional inspection methods are slow, siloed, and dangerously incomplete.

Insulation removal:

  • Disruptive and expensive.
  • Labor-intensive and time-consuming, with a high risk of process upsets and insulation damage.
  • Limited coverage. Often targets a small percentage of piping, leaving large areas unchecked.
  • Health risks: Exposes workers to hazardous materials such as asbestos or fiberglass.

Rope access and scaffolding:

  • Safety hazards. Falls from height remain a leading cause of injury.
  • Restricted time and access. Weather, fatigue, and complex layouts limit coverage and effectiveness.
  • High coordination costs. Multiple contractors, complex scheduling, and oversight, which require continuous monitoring, documentation, and compliance assurance across vendors and protocols drive up costs.

Spot checks:

  • Low detection probability. Random sampling often fails to detect localized corrosion.
  • Data gaps. Paper records and inconsistent methods hinder lifecycle asset planning.
  • Reactive, not proactive: Problems are often discovered late after damage has already occurred.

A smarter way forward

While traditional NDT methods for CUI like Pulsed Eddy Current (PEC) and Real-Time Radiography (RTR) remain valuable, the addition of robotic systems, sensors, and AI are transforming CUI inspection.

Robotic systems, sensors, and AI are reshaping how CUI inspections are conducted, reducing reliance on manual labor and enabling broader, data-rich asset visibility for better planning and decision-making.

ARIX Technologies, for example, introduced pipe-climbing robotic systems capable of full-coverage inspections of insulated pipes without the need for insulation removal. Venus, ARIX’s pipe-climbing robot, delivers full 360° CUI data across both vertical and horizontal pipe circuits — without magnets, scaffolding, or insulation removal. It captures high-resolution visuals and Pulsed Eddy Current (PEC) data simultaneously, allowing operators to review inspection video and analyze corrosion insights in one integrated workflow. This streamlines data collection, speeds up analysis, and keeps personnel out of hazardous zones — making inspections faster, safer, and far more actionable.

These integrated technology platforms are driving measurable gains:

  • Autonomous grid scanning: Delivers structured, repeatable coverage across pipe surfaces for greater inspection consistency.
  • Integrated inspection portal: Combines PEC, RTR, and video into a unified 3D visualization, streamlining analysis across inspection teams.
  • Actionable insights: Enables more confident planning and risk forecasting through digital, shareable data—not siloed or static.

Real-world results

Petromax Refining adopted ARIX’s robotic inspection systems to modernize its CUI inspections, and its results were substantial and measurable:

  • Inspection time dropped from nine months to 39 days.
  • Costs were cut by 63% compared to traditional methods.
  • Scaffolding was minimized 99%, reducing hazardous risks and labor demands.
  • Data accuracy improved, supporting more innovative maintenance planning.

Why the time is now

Energy operators face mounting pressure from all sides: aging infrastructure, constrained budgets, rising safety risks, and growing ESG expectations.

In the U.S., downstream operators are increasingly piloting drone and crawler solutions to automate inspection rounds in refineries, tank farms, and pipelines. Over 92% of oil and gas companies report that they are investing in AI or robotic technologies or have plans to invest soon to modernize operations.

The tools are here. The data is here. Smarter inspection is no longer aspirational — it’s operational. The case has been made. Petromax and others are showing what’s possible. Smarter inspection is no longer a leap but a step forward.

---

Tyler Flanagan is director of service & operations at Houston-based ARIX Technologies.


Trending News