Self-Driving Trucks Could Alleviate Supply Chain Problems

Computer Modeling Helps Build the Future of Autonomous Semi Trucks

Across the globe, hundreds of thousands of workers rise from their beds in the middle of the night to climb into the tight quarters of the cabin of a big rig. This workforce hauls billions of tons of cargo every year to warehouses, groceries stores, and ports. The shipping industry moves nearly 11 billion tonnes (12 billion tons) of freight via human-driven big rigs annually.

With autonomous trucks, TuSimple can make highways safer for truckers and other drivers alike, all while saving money on the transport of crucial goods.

While these rigs are essential to the movement of supply chains around the world, the industry faces a rising driver shortage. The trucking industry has struggled to both retain its existing workforce and certify new drivers, especially as driving schools closed during the COVID-19 pandemic. These circumstances combined have made it difficult for trucking to meet rising demands for supplies ranging from fuel to electronics.

Fleet of trucks parked inside a large garage.

TuSimple fleet of autonomous trucks. (Image credit: TuSimple)

Long-haul big rigs are also a worker safety nightmare. With long hours, often driving through the night or early morning, truck drivers are more susceptible to fatigue than other drivers. The CDC says fatigued driving impairs driving as much as driving while intoxicated.

It was in this combination of problems facing the trucking industry that Xiaoling Han and colleagues at TuSimple saw the opportunity for a new kind of solution: autonomous technology for heavy-duty trucks.

Han, TuSimple’s senior director of sensors and vehicle control integration, says these vehicles could solve a huge problem for the trucking industry. “Autonomous trucks offer a much stronger business case than autonomous cars because the routes are highly-repeatable and the majority of the driving takes place over the highway,” says Han. “While solving the autonomous driving problem for trucks is extremely difficult, it has far fewer problems to solve than autonomous cars.”

With autonomous trucks, Han says TuSimple can make highways safer and more efficient while helping reduce costs and the environmental footprint of trucks.

The Road to Autonomy

It’s not uncommon to see technology across a wide range of industries labeled as autonomous these days, from robot vacuums to your new car. But when it comes to true autonomy, many of these technologies fall short, explains Govind Malleichervu, automotive industry manager at MathWorks.

Level 4 or 5 autonomous trucks would give drivers an opportunity to rest while driving through the night on empty roads.

As defined by the Society of Automotive Engineers, there are six possible levels of autonomy for automotive vehicles, ranging from zero to five. Levels 0 through 3 range from no automation to limited automation, including assisted steering or braking. Due to the limited capabilities, human drivers still control the vehicle and supervise these systems. Level 5 autonomy, where the vehicle performs all driving tasks, is still many years away.

6 levels of autonomous driving from 0 – no automation to 5 – full automation

The six levels of autonomous driving (Image credit: NHTSA.gov)

Trust in the Model

TuSimple is already testing its big rigs on real highways, including a 951-mile trek from Arizona to Oklahoma to deliver watermelons. However, the costs of relying entirely on-road testing can add up quickly, Han explains. Instead, the team largely relies on modeling and simulations as a safe and cost-effective alternative for the by-wire control development. According to Han, as much as 90% of the trucks’ vehicle control unit testing is done using models developed in MATLAB^®, Simulink^®, and other software.

By using models, the team references a single source of truth, automatically generates code, and can test the autonomous systems with different vehicle constraints.

“For the by-wire braking, we only test at the vehicle level 10% of the time, because it’s very cost prohibitive,” says Han. “It consumes human power and requires additional coordination. So, we do most of the testing in simulation.”

The simulations require modeling how different sensor inputs, such as lidar and radar, are relayed through microprocessors to the truck’s physical control systems. Managing these components through a model enables the team members to work on the design simultaneously, even when they’re working from different physical locations.

A TuSimple driver lets autonomous driving take over. (Video credit: TuSimple)

“The team can refer to the model to check what’s going on inside the vehicle,” Han says. “It’s much easier than reading the code. That’s why the model is the single source of truth for the whole Vehicle Control Unit team, because we can communicate based on the model across all related functional teams.”

The models also help reduce errors introduced by human coders by automatically generating code to implement designs made at the model levels, says Malleichervu. This approach enables the team to test the autonomous systems with different vehicle constraints, such as different engines or cargo weights. The team relies on MATLAB scripts and GitHub to manage the variants.

Breaking the Mold

Beyond simplicity, Han says that relying primarily on virtual models to test their new designs also helps them break free of a standard design model across the automotive and software industry called the V-model.

“We have a very fast iteration period, which means we release software in a month not a year. … Without modeling, it’s very hard to release a feature in a month, because in that month you need to finish the coding, testing, validation, everything.”
Xiaoling Han, senior director of sensors and vehicle control integration at TuSimple

TuSimple combines V-model and Agile methodology to create their ideal design process. (Image credit: TuSimple)

Also known as the verification and validation model, the V-model is a design approach that moves systematically through analysis, design, and validation before implementing new features. As a result, Malleichervu says that taking an engine or vehicle program from start to finish with the V-model can typically take three to five years; implementing upgrades to an existing design requires approximately one year.

For TuSimple, this methodology was too slow, Han says. Instead, the company implemented a mixed method that includes both the V-model and Agile methods when developing their autonomous systems.

“We have a very fast iteration period, which means we release by-wire control software in weeks rather than years,” says Han. “That’s why we need to follow the Agile process as well. That’s also why we rely on modeling. Without modeling, it’s very hard to release a feature in a month, because in that month you need to finish the coding, testing, validation, everything.”

Using this hybrid approach, TuSimple can release by-wire software patches in less than 24 hours and new features anywhere between 72 hours and two weeks. It’s this novel design method and TuSimple’s use of modeling that Han thinks will help the company reach its goal of hitting the highways with autonomous big rigs by 2024.

Self-Driving Trucks Could Alleviate Supply Chain Problems

Computer Modeling Helps Build the Future of Autonomous Semi Trucks

The Road to Autonomy

Trust in the Model

Breaking the Mold

Read Other Stories

Autonomous Electric Tractor Brings Artificial Intelligence to the Field

Seeing the Road Ahead

Developing Autonomous Parking Technology for Truck and Trailer Combinations with Model-Based Design