(traffic buzzing)

- [Narrator] February 27th, 2021,

police body cam footage

shows a traffic stop

on a highway in Montgomery County, Texas.

- [Police officer] Hey, how's it going?

- [Narrator] As this 2019 Tesla Model X

driving in autopilot mode,

strikes their vehicles at 54 miles per hour,

injuring five officers,

and hospitalizing the subject of the original traffic stop.

- [Paramedic] Again, roll up, roll up.

One, one, two, three. (patient moaning)

- [Narrator] The crash is one of 16

between Tesla's and emergency vehicles

being investigated

by the National Highway Traffic Safety Administration

or NHTSA, to determine whether Tesla's Autopilot

has contributed to the accidents.

The journal obtained exclusive dashcam footage

and partial data logs from the Tesla in the Texas crash.

We annotated the footage for clarity.

These materials show the car's Autopilot System

failed to recognize the stopped emergency vehicles in time,

and though, its Driver Monitoring System

appears to have worked as designed,

it was not enough to sideline the impaired driver

and prevent the collision.

We also obtained eight Crash Reports

included in the NHTSA investigation.

In at least six,

the incidents occurred

when emergency vehicle lights were flashing.

NHTSA declined to comment on an ongoing investigation,

and Tesla did not respond to a request for comment.

(dramatic music)

The driver in the Texas crash begins his 45-minute trip

just before 12:30 AM.

"He is intoxicated," according to police reports.

The police investigation notes,

"Several instances when he swerves in his lane."

About four minutes into his drive,

the log show he sets the car on Autopilot

at a speed of 64 miles per hour.

Tesla's Autopilot is a system

that partially automates many driving tasks on highways,

including steering, braking, and lane changes.

Drivers using Autopilot are supposed to remain engaged

so they can take control of the car at any time.

Federal investigators have said, "Tesla's marketing,

including the name Autopilot, exaggerates its capabilities,

and encourages drivers to misuse the technology."

The Autopilot System for a 2019 Model X,

the model in the Texas crash,

judged whether the driver was alert

based on whether their hands were on the wheel.

If the system did not detect hands on the wheel,

the driver received an alert.

If the driver didn't respond,

Autopilot would disengage.

"The driver in the Texas crash

receives one of these alerts less than two minutes

after engaging Autopilot,"

according to the car's logs, and he complies.

He receives two more in the next minute, and complies.

In fact, he receives and complies with 150 of these alerts

over the course of this 45-minute drive.

By the design of Tesla's Driver Monitoring System,

the driver was paying enough attention

to operate the vehicle in Autopilot.

This year and model of Tesla

uses a combination of radar and camera technology

to recognize objects in all directions.

Autonomous vehicle experts say,

"The radar can easily recognize moving vehicles,

but that it has difficulty

distinguishing stationary obstacles,

leaving it mainly up to the cameras to detect them."

Around 15 minutes into the drive,

the logs indicate the technology is working.

It recognizes a vehicle about 120 yards ahead.

Then again, about 35 minutes into the drive,

the Tesla sees a vehicle as it merges 70 yards ahead

and tracks it as it drives off.

About 45 minutes in,

the Tesla approaches emergency vehicles

on the side of the road.

Autopilot is not designed to identify them as obstacles

because they're not in a lane,

but an attentive driver

would typically know to slow down or change lanes.

Around the same time,

the driver receives his 150th warning

to keep his hands on the wheel.

He complies, but it's not enough

to get him to respond to the stopped vehicles.

Seconds later, other police cars

are visibly blocking the lane,

but the logs show no sign that the Tesla sees them.

Experts in autonomous vehicle safety

who reviewed the crash footage

say, "There's a difference

between the way the car's camera sees an ordinary vehicle

and an emergency vehicle.

The police car's flashing lights created a hazy image

that the car's software likely did not recognize

as a stopped vehicle."

The logs indicate

that, "The car finally recognized something in its path,

just 2.5 seconds and 37 yards before the crash.

Autopilot attempts to slow the car,

and ultimately, disengages,

but at 54 miles per hour, it is too late.

The five officers injured in the crash

are suing Tesla,

claiming that "The Autopilot Feature

was responsible for the accident."

An attorney for the officers declined to comment.

Tesla denies the lawsuit's allegations,

and claims, "The fault lies with the driver."

The driver did not respond to attempts to contact him.

In 2021, after repeated recommendations

from federal investigators,

Tesla began using internal cameras

to monitor driver attentiveness,

but safety experts continue to find flaws in its design,

and drivers still find ways to fool the system.

The same year,

Tesla also issued a software update

that was designed to improve Autopilot's ability

to detect emergency vehicles.

Though one of the crashes NHTSA is investigating

occurred after these updates.

The government has now expanded its investigation

beyond crashes with emergency vehicles

and is investigating the overall effectiveness

of Tesla's Autopilot System.

And they could ask the company to recall the cars

if they find the technology flawed.