Self-driving trucking is following a parallel track to autonomous passenger vehicles.
Similar technologies are key to making the unmanned, long-haul big rig a possibility. Like those self-driving cars1 already on the road today, autonomous tractor trailers are decked out in an array of sensors: cameras, radar, LIDAR, ultrasonic sensors, and more (depending on the implementation), which all work together to create a living, 3D representation of the world around them. Because the truck is moving at freeway speeds, this simulation has to be updated in real time, multiple times every second.
Meanwhile, powerful computers are necessary to process every data point on that simulated map of the environment. Artificial intelligence algorithms are key, and they represent some of the most prized intellectual property owned by autonomous trucking companies. It’s here that the secret sauce comes to bear on the challenges of moving an 80,000-pound vehicle without a driver in it.
A typical truck will have dozens of algorithms running simultaneously while it’s in motion, each one tasked with identifying objects on the road, determining the best speed and the ideal lane in which to travel, and using a branch of mathematics known as probabilistic modeling (which incorporates random variables into event models and gives probability distribution as a solution) to try to predict as far in advance as possible how other vehicles are going to behave.
One of the leading companies making this possible is TuSimple2, an autonomous trucking pioneer which has been piloting live, autonomous trucking runs on Arizona’s highways since early 2018. The company has made huge strides in a short amount of time, to the point where it actually accepts commercial freight for largely autonomous delivery across the 100+ miles between Phoenix and Tucson. It even earns revenue while it trains its trucks — using edge computing and massive amounts of onboard processing and data storage, rather than the cloud.
Why the Cloud is Not the Answer
Think the cloud can handle the intense processing needs of an autonomous truck? Think again.
“Because we need low latency, we have to run everything locally,” says Xiaodi Hou, TuSimple’s CEO.
There’s simply no time to spend even the few seconds needed to wirelessly send data to a cloud-based server, wait for it to crunch the numbers, then receive a response. By the time the answer came through, it would be too late to matter, because everything in the simulation would have moved hundreds of feet.
Instead of relying on remote servers, self-driving trucks have to carry with them immensely powerful computers, plus plenty of storage, to crunch all this information. The probability models required to make intelligent choices about where to position the vehicle are complex, to the point where a single self-driving vehicle can generate an estimated 4 terabytes of data3 for each full hour of driving. In a big rig, this is compounded because of the sheer size of the vehicle: It’s arguably an order of magnitude more difficult to pilot a 48-foot tractor trailer than a 15-foot long passenger vehicle due to the amount of space the semi consumes.
All told, it’s a massive undertaking that requires a huge amount of processing power and lots of fast, local storage, to the point where TuSimple calls its onboard units “supercomputers.” These powerful, purpose-built computers run upwards of 100 different algorithmic modules, including many that revolve around machine learning and deep learning, that handle everything from processing LIDAR data to tracking moving objects to estimating the speed of other vehicles on the road. There’s also the not-small matter of actually controlling the truck’s throttle, brake and steering, which requires the involvement of a wholly separate engineering discipline.
Only after a successful delivery is the data offloaded from the truck and uploaded to the cloud, where it is processed, algorithms are refined, code is tested repeatedly to meet safety requirements, and everything is redeployed to the truck for the next run.
Driving for Full Autonomy
TuSimple’s trucks today carry a backup driver that can step in if something goes awry, but the goal is to achieve complete autonomy – with no human in the cab – by the end of 2020. Achieving that goal could power a monumental shift in the economics of trucking4. The trucking industry is a complicated one, and it’s experiencing a perfect storm that’s marked by skyrocketing demand (thanks, online shopping) and a shortage of willing drivers that worsens daily. A trucking industry association estimates that 60,000 trucking jobs are currently unfilled5, and that number could hit 100,000 within a few years. Not to mention that trucker mortality rates have went up since 20116.
AV technology in trucking could reduce operating costs by about 45 percent and save the industry over $80 billion annually4 – and possibly help save lives each year, too.
“This is a natural evolution of human civilization,” says Hou. “I’m proud to be on the forefront of it.”
This content is produced by WIRED Brand Lab in collaboration with Western Digital Corporation.
- How Do Self-Driving Cars See?
- Building Out an Autonomous Vehicle Ecosystem *and based on internal estimates
- Distraction or Disruption? Autonomous Trucks Gain Ground in US Logistics
- Facing A Critical Shortage Of Drivers, The Trucking Industry Is Changing
- Truckers Died in Record Numbers on the Job in 2017