TuSimple is working to develop self-driving trucks that will make highways safer and more efficient while helping to reduce transportation costs and their environmental impact.
Every day, hundreds of thousands of drivers around the world get out of bed in the middle of the night and burrow into the cramped driving space of large trucks. This workforce ships billions of tons of goods each year to warehouses, grocery stores, and ports. The transportation industry transports nearly 11 billion metric tons (12 billion tons) of cargo annually through large, manually driven trucks.
With self-driving trucks, TuSimple can turn highways into safer places for truck drivers and others. At the same time, the use of self-driving trucks can also save the cost of transporting critical materials.
While these trucks are critical to the functioning of supply chains around the world, the industry still faces a growing shortage of drivers. The trucking industry has been faced with the dilemma of "retaining old drivers and developing new drivers", which has become even more difficult with the closure of many driving schools during the COVID-19 pandemic. As a result of the combination of these factors, truck capacity has been unable to meet the growing demand for transportation, and the supply of everything from fuel fuels to electronics has been affected.
TuSimple's fleet of self-driving trucks. (Image ownership: TuSimple)
In addition, long-distance transport trucks are also a nightmare for driver safety. Due to long working hours, often driving at night or early in the morning, truck drivers are more likely to develop fatigue than other drivers. The CDC says that fatigue driving is as risky as traffic posed by drunk driving.
It is these problems facing the truck industry that allows TuSimple's Han Xiaoling and colleagues to see an opportunity for a new solution: autonomous driving technology for heavy-duty trucks.
Han Xiaoling, senior director of sensor and vehicle control integration at TuSimple, said these vehicles can solve the tough problems facing the trucking industry. "The commercial application prospects for self-driving trucks are broader than those of autonomous passenger cars, because trucking routes are highly repeatable and most driving takes place on highways," Han said. Although the problem of truck self-driving is very difficult to solve, it needs to face much less problems than self-driving cars. ”
Han Xiaoling also said that TuSimple's self-driving trucks can make highways safer and more efficient while helping to reduce transportation costs and their impact on the environment.
The road to autonomous driving
Govind Malleichervu, automotive manager at MathWorks, said that all walks of life are currently advertising autonomous technologies, from intelligent sweeping robots to various new cars; however, if measured by true autonomy standards, many technologies do not achieve strict autonomy.
Level 4 or 5 self-driving trucks give their drivers a chance to rest while driving on open roads at night.
According to the International Society of Automata Engineers, the autonomy of a car consists of six possible levels from 0 to 5. Levels 0 to 3 range from no automation to limited automation, including auxiliary steering or braking. Due to functional limitations, human drivers still need to control the vehicle and supervise these systems. Achieving Level 5 autonomy, where the vehicle performs all driving tasks independently, will still take years.
Six levels of autonomous driving (image ownership: NHTSA.gov)
Malleichervu said, "Many OEMs offer self-driving capabilities that are Level 2 or Level 2+. The plus sign means increased autonomy on a level 2 basis, but not yet level 3. At this level, the Advanced Driver Assistance System (ADAS) on the vehicle can control the braking system, acceleration and steering, while the human driver remains responsible for monitoring the environment at all times and performing all other driving tasks. ”
Part of the reason for this relatively limited autonomy is the safety concerns that accompany testing and developing vehicle ADAS in urban environments.
Han Xiaoling said that this is where the Advantage of TuSimple Trucks lies. While the company expects its trucks to be fully on the road until at least 2024, testing the technology on less-intensive highways rather than downtown offers an easier path to fully autonomous driving. In order to test self-driving cars in cities, companies must first reach an agreement with local authorities on the area, frequency, and speed of the vehicles to be tested. This often means that there will be less area and time available for testing.
Another benefit of highway testing is that it faces fewer pedestrian barriers than urban roads and largely avoids stop-and-go traffic conditions and narrow or uneven roads. This means that in case something goes wrong, while the risk of injury to other drivers is not zero, it does reduce the likelihood of a fatal accident.
Han Xiaoling said that TuSimple's trucks have now reached Level 4 autonomy. This means that vehicles can drive themselves autonomously under limited conditions without the help of a human driver — although TuSimple still plans to have human drivers monitored in the driver's seat. Current trucking requires human drivers to be vigilant at all times, and the difference between Level 4 or Level 5 autonomous trucks is that truck drivers have a chance to rest when driving all night on open roads.
Han Xiaoling and his team are developing one of the key technologies for this kind of autonomous driving: autonomous line control motion systems. The weight of large trucks and the dynamics that come with towing a fully loaded trailer make it more difficult for trucks to stop quickly. A fully loaded truck weighs 36,000 kg (80,000 lbs), while a single bus weighs only 1,360 kg (3,000 lbs). The braking system must be able to safely block the forward thrust of heavy trucks at high speeds.
Han Xiaoling said, "The line control dynamic system is very complex, the entire system pipeline starts from perception, and our program should observe the situation of the vehicle front and back. This involves motion planning, prediction, and control algorithms. ”
With sensors such as perception and lidar, TuSimple's 4-level autonomous system provides a 360-degree view of the vehicle's surroundings and the 1,000-meter range ahead. This data is then passed on to the safety-critical line control dynamic control system, Han Xiaoling explains. The control system then establishes an association between software perception and the truck hardware to activate safety braking.
Han Xiaoling also said, "Braking is the most basic system in automatic driving, and the requirements for brake safety are higher than all other systems. ”
These stricter standards for braking systems are also reflected in their redundancy. This means that if something goes wrong, the human controller can intervene and apply the brakes manually.
"For other components of the vehicle, such as the engine, a certain degree of redundancy is sufficient, but since the braking system is a core component of vehicle safety, we must have full redundancy, including physical, signal, power and software systems." Han Xiaoling continued.
Trust model
TuSimple has now tested its large trucks on real highways, including a 951-mile journey from Arizona to Oklahoma. However, Han Xiaoling explained that relying entirely on road tests will lead to rapid cost increases. Instead, the team now relies heavily on modeling and simulation as a cost-effective and secure alternative to by-wire development. According to Han Xiaoling, up to 90 percent of truck vehicle control units are tested using models developed by MATLAB®, Simulink®, and other software.
By using models, the team references only unique data sources, automatically generates code, and can test autonomous systems with different vehicle constraints.
"We only control the movement of the test line at the vehicle level 10% of the time because it is very expensive to do so," says Han Xiaoling. "Not only is it labor-intensive, but it also requires additional coordination. Therefore, we use simulation for most of our tests. ”
Simulation requires modeling how different sensor inputs, such as lidar and radar, are relayed to the truck's physical control system through a microprocessor. Using models to manage these components allows team members to design at the same time, even if they work in different physical locations.
The TuSimple driver let the car drive itself. (Video ownership: TuSimple)
"Team members can use models to check what's going on inside the car," Says Mr. Han. "It's much easier than reading the code. That's why models are the only source of data for the entire vehicle control unit team, as all of our relevant functional teams can communicate based on the model. ”
Malleichervu says these models can also help reduce errors introduced by programmers by automatically generating code to implement designs made at the model level. This approach allowed the team to test autonomous driving systems with different vehicle constraints, such as different engines or cargo weights. The team relies on MATLAB scripts and GitHub to manage variations.
Get out of the way
In addition to being simple and easy to use, relying on virtual models for most new design testing helps them stop relying on the standard design model known in the automotive and software industry as V-models, Han said.
"Our iteration cycle is very fast, which means that the release of our software is monthly rather than annual. Release software within a month instead of a year. ...... Without modeling, it is very difficult to release a feature in a month, because programming, testing, verification, and so on all need to be done within this time period. ”
Xiaoling Han, Senior Director of Sensor and Vehicle Control Integration at TuSimple
TuSimple combines the V model and agile methodologies to create the ideal design flow. (Image ownership: TuSimple)
The V-model, also known as the Validation and Validation Model, is a design method that systematically analyzes, designs, and validates new features before they are implemented. Malleichervu says it typically takes three to five years to complete the entire design of an engine or vehicle project using the V-model design approach; it takes about a year to upgrade an existing design.
Han Xiaoling said that for TuSimple, this method is too slow. Instead, the company adopted a hybrid approach to developing autonomous driving systems, using a combination of the V model and agile methods.
"Our iteration cycle is very fast, which means we have to release the software by wire in a matter of weeks instead of years," Han xiaoling said. "That's why we also need to follow agile processes. This is why we rely on modeling. Without modeling, it's hard to release a feature in a month because programming, testing, validation, and everything else needs to be done within that month. ”
Using this hybrid approach, TuSimple can release patches for the software-by-wire within 24 hours and new features within 72 hours to two weeks. Han Xiaoling believes that it is this novel design approach and TuSimple's use of modeling that will help the company achieve its goal of driving large trucks on the highway by 2024.