TUCSON, ARIZONA (AP) – According to insurance industry figures, nearly seven out of ten car accidents happen within ten miles of home.
It appears to be true even for self-driving trucks.
On a sunny Wednesday morning last month, an otherwise unremarkable ride in TuSimple’s high autonomy-equipped International LT became dangerous when four eager vehicles raced past the truck on the left as it began a wide right turn into the corporate driveway.
Then, on the right, a white Jeep Compass sped by, infringing on the bike lane to finish its maneuver. Randy Redstone, the safety driver, slammed on the brakes. He believes the robot driver would have come to a halt if he hadn’t intervened. Despite the fact that the truck’s array of cameras had picked up the oncoming Jeep, Redstone couldn’t risk it.
Redstone, who had driven heavy rigs for 35 years before joining TuSimple three years ago, said, “I’ve taken that turn hundreds of times, and that was the worst I’ve ever seen.”
Even the most advanced robot trucks must be prepared for silly human pranks, as demonstrated by a recent near-miss on Old Vail Road, a two-lane blacktop in the Sonoran Desert approximately 15 miles east of the city.
Chuck Price, TuSimple’s chief product officer, who sat across from me in the sleeper cab shared with the file cabinet-sized compute unit, said, “That was about the most nutso man I’ve ever seen.” “That guy was really insane.”
Within 30 minutes, TuSimple’s 360-degree cameras recorded and replayed the close call for analysis. While a perfect journey would include no disengagement of the autonomous system, an example like this can teach us a lot.
Price explained, “We can pull that data very quickly and deliver it into the hands of engineers very quickly.”
A breakup isn’t something to be ashamed of. The driver monitors and engineers in the front of the cab have only one job: to ensure that the vehicle operates safely. In 5.4 million miles of autonomous driving, TuSimple has had no at-fault collisions.
Drivers are the most knowledgeable.
This is partly due to the fact that its 67 safety drivers have just as much influence in how the trucks learn as physical tests and computer simulations. Some of the company’s drivers were employed after completing classes at Pima Community College. Others are retirees who have logged millions of miles without an accident.
“Driving best practices are part of what we automate,” Price said. “The driver team agrees on a collection of circumstances that happened that week or in a previous couple of weeks.” They generally choose circumstances in which the drivers disagree. And we get them to agree on what the appropriate behavior is.”
TuSimple’s product, engineering, algorithm, and systems teams capture this information and incorporate it into the autonomous system, promoting AI and machine learning. They can also function with black box-style recordings of the driver and engineer.
“If a triage team trying to figure out what happened on a run is interested, they can simply put on headphones and listen to what they were saying.”
“We’ll catch it if they bring us a new understanding,” Price said. “Or we’ll talk about it if there’s an existing necessity that conflicts with that.” And then we’ll say, ‘OK, we’ll have to change our demand.'”
Consider two locations near TuSimple’s campus on Rita Road, just off Interstate 10. Human driver expertise was input into a machine that does 600 trillion calculations per second, not a typo.
One is near a University of Arizona technology park entrance with bike lanes and “a lot of pedestrian congestion,” according to Price. On that stretch, TuSimple trucks decelerate to 35 mph.
The other is a slightly disguised road on the east side of Rita Road that leads to a Pilot Express gas station with a Subway and a Burger King.
“That petrol station has a variety of problems,” Price stated. “It’s common for people to cut in front of trucks. So, for reasons I’m not sure of, this truck is always a little reluctant around there.”
Freight transport speed
On I-10, however, the majority of the westbound road to Tucson is two or three lanes wide. The robot truck is forceful, implying a future in which drivers are no longer needed and all decisions are made by the robot.
The TuSimple truck usually took the center lane rather than the right lane in three-lane sections. When the opportunity arose, it signaled and moved into the left lane to pass.
“It’s best if you have as much knowledge as possible.” So we may calculate lane performance and say, ‘Actually, it’ll be better to get across.’ That lane appears to be becoming busy. Let’s get over here early so we don’t have to slow down,'” says the narrator. Price stated his opinion. “We want to keep the mass moving at the same rate as much as possible.” We don’t want to be slowed down [or] sped up a lot.”
UPS is testing routes with safety drivers and engineers from Arizona to Orlando, Florida, and Charlotte, North Carolina – a practically coast-to-coast autonomous network. Before the end of the year, it plans to test removing the safety driver and engineer from the cab for multiple runs over several weeks in Arizona.
TuSimple’s goal is to have robot drivers working 20 hours a day to address a driver shortage, particularly in the long haul, when turnover is about 100%. A purpose-built Navistar Class 8 LT variant has roughly 7,000 nonbinding bookings. Its present fleet of 58 vehicles is made up of Peterbilt and Navistar trucks that have been upgraded.
Voices can be heard
Every motion is aired while safety drivers and engineers are in the cab. “Intersection clear,” one of the notifications reads. “The speed limit has been raised to 25 miles per hour.” “Before the shift to the left.” “It’s time for a change.” “Left slant.” “Right-wing slant.” The occasional beep indicated that two messages were overlapping, and “it filtered itself,” according to Price.
The TuSimple truck’s reaction to another vehicle approaching it is called bias.
“I used to hear a voice say serve.'” “I didn’t like it,” Price expressed his displeasure.
The voice usually validates what Redstone already knows, and it’s a step forward from the beginning.
“When we first started doing this in 2017, we learned a lesson.” Because they see intent on their display, the right seater [safety engineer] would become much more animated and say, ‘It’s going to change lanes.’ ‘It’s going to change lanes!’ they’d scream at the driver. It’s about to switch lanes! It’s about to switch lanes! Prepare yourself!
“And we remarked to one other, ‘You know what? This is not a sound business strategy.’ These [safety drivers] were suffering from heart attacks. ‘OK, we’ll just have the [automated voice] talk to the driver,’ we decided. ‘Relieve the stress.’
