United States TuSimple
TuSimple is a fast-growing startup dedicated to providing the tools necessary for a level 4 autonomous truck. The company was founded in 2015 by Dr Xiaodi Hou, a world leader in computer vision software from the California Institute of Technology.
The company has facilities in China, California and Arizona, and has so far raised $82 million in financing rounds (2018).
The goal of TuSimple is to provide the tools to transform an ordinary truck into a level 4 autonomous truck. Today it retrofits the technology to combustion trucks in collaboration with Peterbilt, an American trucking OEM, rather than developing it solely for an electric truck.
"We don't think batteries are there yet in terms of the range they can provide" was a comment from Robert Brown, Director of Public Affairs.
Indeed, at a minimum, the freight industry needs around 500 miles for a haulage truck for intercity travel - a target only Tesla has claimed for its Semi, due to start production in 2020.
TuSimple has opted to use a 360-degree viewable camera system over using Lidar to map the truck's local environment - one of the few in the race for autonomy to join Tesla in doing this. While camera-based systems are currently cheaper than Lidar-based systems, for a haulage truck it is not just about lowering the cost, according to Brown.
"An 80,000-pound truck might have a stopping distance of 100 metres, which is already close to the range limit of a Lidar-based system. A camera system can go up to 300 metres, which results in the extra safety for the truck."
In other words, due to the extra weight, the stopping distance is longer and the viewing range always needs to be longer than the stopping distance (otherwise a truck could identify an obstruction or other issue but would crash into it anyway).
Another reason is that the Lidar systems are fragile: TuSimple is targeting and building its systems on combustion trucks which have powerful vibrations and could easily break Lidar. A camera system is more robust because it is solid state. TuSimple is technology agnostic so could end up using Lidar in the future provided costs fall, although this is more likely to be used to enhance a camera system rather than replace it due to the longer stopping distances needed for trucks.
IDTechEx was told the cost of the system to OEMs including 10 cameras is roughly $20,000 - currently lower than that of a Lidar system, which in R&D scenarios has been as high as $90,000.
One issue is that the hardware for autonomous driving is very energy intensive. TuSimple's choice of combustion trucks from Peterbilt requires an additional alternator (two in total) to handle the extra 2kW required of its standard 10 camera system.
Road tests are done in Arizona as the home state of California has a regulatory barrier in place. California has mainly focused on autonomous cars in the past, but now is drafting legislation to include trucks and freight vehicles. Regulation is one of the key barriers to autonomous driving, and today (June 2018) in the US it works on a state by state level. The company is working with the US Department of Transportation to help policymakers develop a federal regulation.
"The technology will be ready before the regulation is: we will have level 4 autonomy by 2020" stated Brown. By contrast, Martin Daum, chief executive of Daimler Trucks, announced in June 2018 it will take at least five years before the driverless technology (level 4) becomes commercially available to its customers - a more cautious prediction.
TuSimple is testing its systems on five Peterbilt trucks in Arizona and is on track to have 25 on the state's roads by the end of the year.
In an interesting comment, Brown stated that it is too dangerous to stop at level 3 autonomy: level 3 still requires a driver's full attention, but the vehicle drives itself so well that the driver tends to get bored at the wheel (and stops paying attention), resulting in an accident. There is a worry that highly publicised accidents will lead to slow consumer acceptance of the technology when in reality it is safer.
When asked about level 5 autonomy, Brown responded:
"Technically a level 5 vehicle is not possible - you would only be able to drive at 'level 5' on specific routes like on highways and within cities, but not anywhere in the world, which is what level 5 means"
With regards to deep learning, Brown commented that most of the code is done the old-fashioned way, and the neural network is trained in a simulated environment. It is useful for learning the nuances of driving culture in different geographies. For example, in China drivers cross over the white lines on the roads at a traffic light in order to give each other space, which you could never train the software to do using a rulebook.
In China TuSimple's trucks are more for terminal trackers than haulage, and the company was granted permission in December 2016 by the government in the Caofeidian District of Tangshan, Hebei Province, China to test on a 60 mile stretch of highway.
So far the company has done 1500 miles of testing in total in the US and China.
"China and the US seem to be travelling at the same pace when it comes to the adoption of autonomous driving technology" stated Brown.
Business Model and Market
Today the company is not yet at a commercial stage and is still in the R&D process. It expects the technology to be ready by 2020. Ultimately the vision is to provide a software-only subscription service to OEMs, who pay a monthly fee to TuSimple. However, in the short term, TuSimple will provide the hardware (outsourced from undisclosed partners) and the software, which can be retrofitted to existing trucks.
The addressable market is all haulage trucks in the US and China.
Cost to OEM
10 Camera System
$82 million (June 2018)
70 full-time employees, and 20 interns over the Summer period (2018)
Knowhow and expertise surrounding computer vision software, as well as two patents in algorithms issued in April 2018.
Not yet profitable
|Product Value Proposition|
|State of Technology/Product Development|
|Medium-Term Commercial Opportunity|
|IP and Know-How|