In the push to develop self-driving technology, most companies have treated public roads as their personal laboratory.
Self-driving startup Ike Robotics Inc. has taken the opposite approach. Though the company has functional prototypes of its system, it is opting not test on public roads for the foreseeable future.
Testing on public roads presents practical complications for enhancing Ike's system, says Alden Woodrow, co-founder of the year-old trucking startup. Whether it is Ike's technology or competitors', he also says public-road testing brings inherent risks that he'd prefer to minimize.
"That's a very different approach and not part of the typical playbook," Woodrow told Automotive News. "But we feel really good about that approach. We don't need to test on public roads to make progress."
Ike detailed its approach to testing and development in a 90-page safety report filed Thursday with NHTSA. The voluntary safety self-assessment describes a process that relies heavily on simulation and private-track testing.
The report is the 16th the federal agency has received from industry players and perhaps the first from a company that has yet to test its technology on public roads.
Starsky Robotics and TuSimple, two self-driving truck competitors, have extensively tested their automated driving systems on public roads. In June, Starsky conducted a test on the Florida Turnpike with no humans aboard. TuSimple hauled U.S. Postal Service packages between Dallas and Phoenix this year.
Testing debate
Other companies have pushed back against the conventional notion that hundreds of test cars must drive millions of miles on public roadways to vet the competence of self-driving systems. Notably, Aurora Innovation has kept its test fleet to a dozen or so cars. In July, a Volvo Cars executive called the idea of measuring progress by miles "a myth."
Ike might be the first to eschew public-road testing altogether.
The company has Class 8 trucks on the roads in California and Arizona, but they're manually driven. The goal is to gather data on scenarios encountered on highways, and then feed that data into simulation, where the company's testing takes place in earnest.
"We run thousands of tests every night on our software to measure its performance, and what we'd like to do is scale that dataset up in substantial ways and move the performance in the right direction as opposed to fixing anecdotal things," Woodrow said. "That's the trap you fall into with public testing. A merge doesn't go well, and you make a fix for that particular situation."
While the company says it will eventually test its technology on public roads, it intends to minimize those efforts.
"We develop meaningful measures of system maturity that do not rely on demonstrations and other anecdotal representations of progress," the safety report says. "We have an enormous responsibility to employ the highest possible safety standards and strict risk-mitigation strategies at every stage of development."
Its name a nod to former President Dwight "Ike" Eisenhower, largely responsible for the development of the interstate highway system, Ike has raised $52 million in funding, according to Crunchbase.
Defaulting to transparency
Beyond its unusual testing approach, Ike has pledged to make information about its fleet and ongoing operations public on its website. Specifically, Woodrow says, the company will share an inventory of all vehicles in its fleet, along with license plate numbers and VINs.
Further, it will maintain a map of all roads where its trucks operate, currently confined to human-driven trips logged on Interstate 10 in Arizona and several interstate spurs and highways in northern California. Ike will also share safety and compliance scores that are maintained by the Department of Transportation and the Federal Motor Carrier Safety Administration, the trucking counterpart to NHTSA.
"We feel like this is appropriate for people to know," Woodrow said. "The public has a right to know as much as possible about what companies like ours are doing. We are developing new technology and testing it in various ways, and others are doing the same. Rather than default to secrecy, what we want to do is default to transparency."
