Washington, DC(CNN) Carmine Cupani wanted to set the record straight. So he asked his young son to stand in the path of his Tesla as it sped through a parking lot.
The North Carolina resident set out to refute a widely circulated video of a Tesla with the company's "full self-driving" beta software -- which allows the car to steer, brake and accelerate, but requires an attentive human driver ready to take the wheel -- plowing into child-size mannequins.
Dan O'Dowd, a software company CEO who published the video earlier this month, thinks the National Highway Traffic Safety Administration should ban "full self-driving" until Tesla CEO Elon Musk "proves it won't mow down children."
That's when Cupani, who runs an auto shop focused on imports and Teslas, got involved and recruited his son. While he's a self-described "BMW guy," Cupani says the software can't compare to what Tesla offers. It also wasn't the first time he'd enlisted his son, who Cupani said is 11 years old, in a potentially viral car endeavor: Earlier this year he posted a video of his son driving his Model S Plaid — which can reach 0-60 in 1.99 seconds — in a private parking lot. It's been viewed more than 250,000 times.
"Some people look at it and say, 'Oh this crazy dad, what is he doing?'" Cupani told CNN Business. "Well, I do a lot of stuff like that, but I'm going to make sure my kid doesn't get hit."
Cupani filmed the test of "full self-driving" in a parking lot. His son stood near the end of an aisle holding a smartphone to film the test. Cupani accelerated the Tesla from the other side of the lot and turned on "full self-driving," reaching 35 mph. The Tesla braked steadily and came to a full stop -- well ahead of his son.
Cupani did another test with his son on a street using Autopilot, the more rudimentary Tesla driver-assist software, and found it stopped for his son, too. "This Dan guy, he says he's an expert in this, expert in that," Cupani said. "Well, I'm an expert in automotive, future technology, pro driving instructor."
Cupani is among the many Tesla supporters who took issue with O'Dowd's video, and set out to create their own tests. Some asked their children to help. Others built homemade mannequins or used blow-up dolls.
The passionate defenses and criticism of "full self-driving" highlight how the technology has become a flashpoint in the industry. The California DMV recently said the name "full self-driving" is deceptive and grounds for suspending or revoking Tesla's license to sell vehicles in the state. Ralph Nader, whose 1960s criticism of the auto industry helped spark the creation of the National Highway Traffic Safety Administration (NHTSA), joined a chorus of critics of "full self-driving" this month.
But it also is yet another example of the unintended consequence of deploying an unfinished, disruptive technology in the wild -- and shows how far some Tesla believers are willing to go to defend it and the company. Enough people appeared to be pursuing their own experiments that one government agency took the extraordinary step of warning people not to use children to test a car's technology.
"Consumers should never attempt to create their own test scenarios or use real people, and especially children to test the performance of vehicle technology," NHTSA said in a statement Wednesday. The agency called this approach "highly dangerous."
Earlier this month, California resident Tad Park saw that another Tesla enthusiast wanted to test "full self-driving" with a child, and volunteered two of his children. Park told CNN Business it was a "little tough" to get his wife to agree. She agreed when he promised to drive the vehicle.
"I'm never going to push the limits because my kids are way more valuable to me than anything," Park said. "I'm not going to risk their lives in any way."
Park's tests, unlike O'Dowd's, started with the Tesla at 0 mph. The Tesla stopped in all of Park's tests ahead of two of his children involved in the video, including a 5-year-old. Park said he wasn't comfortable doing a higher speed test of 40 mph -- like O'Dowd performed using the mannequins -- with his kids.
Toronto resident Franklin Cadamuro created a "box boy," a childlike form crafted from old Amazon cardboard boxes. "Do not blame me for what the car does or doesn't do," he posted at the beginning of his video. "I am a big Tesla Fan."
His Tesla slowed as it approached "box boy." Then it sped up again and struck his cardboard mannequin. Cadamuro speculated that this could be because the cameras could not see the short boxes once they were immediately in front of the bumper, and therefore forget they were there.
Human babies learn at about eight months that an object out of sight still exists, many years before qualifying for a driver's license. But the ability may still elude some artificial intelligence systems like Tesla's "full self-driving." Another Tesla fan found a similar result.
Cadamuro said his video started as entertainment. But he wanted people to see that "full self-driving" isn't perfect.
"I find a lot of people have two extreme thoughts of the 'full self-driving' beta," Cadamuro said. "People like Dan think it's the worst thing in the world. I know some friends who think it's near perfect."
Cadamuro said he also performed other tests in which his Tesla, traveling at higher speeds, effectively steered around "box boy."
Detecting smaller objects like young children quickly and accurately will generally be more difficult than sensing large objects and adults for a computer vision system like what Tesla vehicles rely on, according to Raj Rajkumar, a Carnegie Mellon University professor who researches autonomous vehicles.
The more pixels an object takes up in a camera image, the more information the system has to detect features and identify the object. The system will also be impacted by the data it is trained on, such as how many images of small children it's exposed to.
"Computer vision with machine learning is not 100% foolproof," Rajkumar said. "Just like diagnosis of a disease, there are always false positives and negatives."
Tesla did not respond to a request for comment and generally does not engage with the professional news media.
Following criticism from Tesla fans of his original tests, O'Dowd released another video this month.
Some Tesla supporters had criticized O'Dowd's use of cones as lane markings in his original testing, which may have limited the sedan's ability to steer around the mannequin. Others claimed that O'Dowd's test driver had forced the Tesla to strike the mannequin by pushing the accelerator, which wasn't visible in videos O'Dowd released. Some Tesla enthusiasts also pointed to a blurry messages on the screen of the Tesla vehicle as an indication that O'Dowd's test driver was pushing the accelerator to rig the tests.
O'Dowd told CNN Business that the blurry messages referred to supercharging being unavailable, and to uneven tire wear. CNN Business could not independently verify what the message said as O'Dowd provided no crisper video of what happened in car during the tests.
In his second video, O'Dowd's tested without cones on a residential street and showed the interior of the Tesla, including the accelerator pedal. The Tesla, as in O'Dowd's other tests, struck the child mannequin.
O'Dowd lamented earlier this year in an interview with CNN Business that no industry testing body examines the code for "full self-driving." The US government has no performance standards for automated driver-assist technology like Autopilot.
O'Dowd is the founder of the Dawn Project, an effort to make computers safe for humanity. He ran as a candidate for the US Senate unsuccessfully this year in a campaign focused exclusively on his critique of "full self-driving."
NHTSA is currently investigating Tesla's driver-assist technology so changes may be ahead.
"The software that controls billions of people's lives in self-driving cars should be the best software ever written," O'Dowd said. "We're using absolute, Wild West chaos rules and we've gotten something that is so terrible."