Now Reading
Chinese company TuSimple to develop self-driving trucks in Tucson

From the archive: This story is more than 5 years old.

Chinese company TuSimple to develop self-driving trucks in Tucson

  • TuSimple's system includes nine video cameras and three sets of millimeter-wave radars, the company said.
    TuSimple TuSimple's system includes nine video cameras and three sets of millimeter-wave radars, the company said.
  • The site of TuSimple's Tucson office.
    Dylan Smith/TucsonSentinel.comThe site of TuSimple's Tucson office.

A nondescript warehouse building on a dead-end road in an industrial park on Tucson's West Side is set to be a starting line for what many industry experts see as the future of transportation: self-driving vehicles.

A Chinese company with more than $20 million in recent backing will bring as many as 100 engineering jobs to Tucson as it develops driverless commercial trucks. Beijing-based TuSimple, which tested a vehicle in Arizona in June, is set to announce Monday that it's setting up shop here.

Rather than focusing on the private market of personal vehicles like Tesla and Google are with their autonomous car programs, TuSimple is working to develop an artificially intelligent truck that can transport freight. Company engineers are developing algorithms and equipment, in partnership with a large Chinese truck manufacturer that the company hasn't identified.

TuSimple representatives, local officials and economic development agencies wouldn't go on there record before the public announcement scheduled for Monday afternoon, but has confirmed that the Chinese company will operate in Tucson.

Public records filed with state and local agencies show that TuSimple will have a site here.

The company also has research and development centers in Beijing and San Diego, and has about 100 employees. The company said it tested its self-driving platform with a run between San Diego and Southern Arizona in June.

From Tech Crunch:

The tech used by TuSimple includes extensive HD mapping of the routes driven, as well as three millimeter-wave radar units, in addition to the camera data from eight cameras and the resulting computer vision processing it does. TuSimple says that it can achieve “centimeter-level” accuracy for truck positioning, even when driving inside a tunnel, and its in-house decision-making machine intelligence makes for safe route navigation.

The company’s also developing a car identification tool using its image recognition software. They note that once available, it’ll be able to identify car make and model with 97 percent accuracy from photos of vehicles uploaded to its site. This is an interesting additional use of its image processing chops, and could present additional revenue opportunities once its self-driving trucks are on roads capturing images of other vehicles on the road while en route.

The company said that "TuSimple team members come from California Institute of Technology, Carnegie Mellon University, the Hong Kong University of Science and Technology, Nanyang Technological University, Tsinghua University, Peking University and Shanghai Jiaotong University. 60 percent of its employees have a Ph.D. degree."

Officials with Sun Corridor, the regional economic development agency, said that TuSimple's plans for Tucson will be unveiled Monday. Scheduled to appear at an announcement of an "important economic development" are Pima County Supervisor Sharon Bronson, Tucson Mayor Jonathan Rothschild, Dave Hutchens of Tucson Electric Power and Sun Corridor CEO Joe Snell.

Officials contacted by declined to comment before the announcement, as did TuSimple representatives.

The company has outlined its broader plans to the industry press, with CTO Xiaodi Hou telling Technology Review last year that intercity freight will be a huge market in China:

TuSimple is collecting data aboard a number of manually driven trucks. The company aims to demonstrate the technology in the first quarter of next year, and to have a commercial demonstration in 2018.

The approach TuSimple is focused on is particularly cost-effective. It relies heavily on computer vision and algorithms that can understand a scene in detail, going beyond identifying vehicles to predicting what actions they may take. “Everything is done in computer vision with deep learning,” Hou says, referring to a popular kind of machine learning that involves feeding large amounts of training data into a big neural network.

Hou, who completed a PhD in computation and neural systems at Caltech under Christof Koch, a renowned neuroscientist now at the Allen Institute for Brain Science, has been working on advanced computer vision for many years.

Across China, around 7.2 million trucks and 16 million drivers are responsible for intercity transportation of goods, according to figures provided by TuSimple. This industry is worth more than $300 billion, and drivers account for around 40 percent of the costs incurred by truck companies. Some long-distance trips across China require two or even three drivers to complete. Autonomy would allow a single driver to sleep during long highway stretches.

"The truck freight industry in the U.S. is even bigger, valued at around $700 billion," Technology Review reported.

Among TuSimple's backers is graphics card giant Nvidia, which announced this week that it had joined in a $20 million investment round with Sina, the company behind Chinese behemoth Weibo.


TuSimple's sensor tech

The company's summary of its technology for self-driving trucks:

TuSimple’s autonomous driving platform uses an array of cameras to scan the surrounding environment. With trucks, the sensors must be even more powerful due to the longer stopping times required. Our platform’s camera and millimeter-wave radar array allows for sensing distances of up to 200 meters from the vehicle, compared to the standard 80 meters using LiDAR arrays. This allows our system to safely observe its surroundings when driving heavy freight trucks.

TuSimple’s original deep learning detection algorithms enable the cameras to perceive the surroundings just like the human eye. It can detect and track objects within your field of vision in real time, and make pixel-level interpretations of the visible scene. With original high-precision visual positioning and multi-sensor integration technology, a truck can achieve a centimeter-level of positioning accuracy – even in a tunnel. TuSimple’s self-developed artificial intelligence decision-making system can even guide vehicles along a safe and fuel-efficient route based on terrain and real road conditions.

— TuSimple

— 30 —

Top headlines

Best in Internet Exploder