During this summer, I worked with TuSimple, one of the leading autonomous truck solution developer, as a business intern, and this is a translation of an article that I wrote when working there. The Chinese version can be found here.
To focus on the key points, I will not elaborate on some general background knowledge such as the potentials of this technology and the background of companies. I will try to spend most of the space sharing with you my insights on the future of autonomous trucks. “Truck” refers to heavy-duty (class 8) trucks specifically in this article.
Market: why should we care about autonomous trucks and who are the big players?
Why should we care about autonomous trucks?
- Huge potential market. The autonomous truck market is estimated to be 270 billion in the United States and 5 trillion in China.
- Clear pain points. There is an increasing shortage of truck drivers in both markets.
- Fast to market. Most people in the autonomous driving industry would agree that this technology will be first commercialized on trucks rather than for passenger cars. The major reasons include:
- the routes are relatively fixed, making it easier to manage the unexpected situations;
- the scenario is relatively less complicated – with few traffic lights, few pedestrians, and other obstacles;
- the end users (shippers/logistics companies) are well educated about the benefits and have high willingness to pay.
What are the existing players?
As of September 2018, the major players (in terms of size of teams and accomplished test miles) in this space include Tesla, Google, TuSimple, and Embark. Other than these companies, there are some other smaller start-ups such Starsky Robotics. The four major players have different positions and strategies when approaching autonomous trucks.
- Tesla – the only OEM among the Big Four. Having the experience and ability to manufacture vehicles is a huge advantage, since the other big players are coming from the software industry with little hardware (vehicle) expertise. Tesla’s OEM experience can help them have a better integration between the AV system and the truck’s control system. Moreover, the collaboration with OEMs has been a bottleneck for the AV companies for a while since most OEMs are hesitating about what type of relationship to have with these disrupters. For the other players, a weak relationship with OEM can limit both their product development and fleet building. Furthermore, Tesla is the only one among the Big Four that is working on electric trucks.
- Google (Waymo) – the only Internet giant among the Big Four. As an ambitious Internet giant, Google is expected to spend most of its autonomous driving resources on passenger cars, the market of which will eventually be much bigger than that of trucks. Moreover, the technology for passenger cars cannot be easily translated for trucks, and this difficulty means that no company is going to win the two markets (passenger car/truck) at the same time. Therefore, Google will primarily compete with Cruise and Zoox on the passenger car markets.
- TuSimple – a 300-employee D-round start-up focusing solely on autonomous trucks. Unlike the first two big players who have their main businesses, TuSimple is solely focusing on autonomous trucks and has the biggest development team (~300 employees in the US and in China) dedicated to the technology. This position enables TuSimple to concentrate all its resources on its only product, and being a start-up also gives the company a lot of flexibility in iterating the technology, updating the business strategies, and finding the best position. Moreover, being the only company with a multi-national background among the Big Four players, TuSimple can leverage the resources (e.g. governmental relationships, partners, talents, customers) in both sides.
- Embark – another start-up focusing solely on autonomous trucks. Similar to TuSimple, Embark is a start-up dedicated to this technology but with a smaller team. The company is young but growing very rapidly. Both Embark and TuSimple have entered the stage of commercial trials, and they are also testing their trucks on I-10. Given the company’s development status, I include Embark as one of the big players.
Technology: differences between autonomous trucks and autonomous cars?
Many investors wonder whether it is easy to transfer the autonomous driving technology from passenger cars to trucks. The short answer is NO.
1. Different use cases and priorities
As a tool for production, trucks are mostly used as a tool for transporting goods, and the priorities for autonomous trucks are: safe, highly efficient, and low-cost. More specifically, the heavy-duty trucks will be mostly operated at a high speed on highways.However, for passenger cars, they are mostly used for daily services. People’s primary expectations on autonomous cars are: safe, comfortable, and value-adding (e.g. people can do more things in the cars). Efficiency is not a big issue here because it relies heavily on the urban traffic conditions. Cost is also not a big issue here since people might not sense the two or five cents per mile cost reduction. For most of the time, cars will be operated at a low-to-middle speed in urban areas (local roads).
Therefore, companies have to cater different priorities when developing the autonomous driving solutions for cars and trucks.
2. Different behavior
Trucks and cars are different in their designs and performance. The biggest difference is that: trucks should not do hard braking frequently and require a much longer brake distance than passenger cars. Other than the economic reasons (e.g. fewer brakes means lower costs), some other reasons support this argument. Comparing to passenger cars,
- Trucks carry different goods. A hard braking will cause severe damages for trucks. Unlike passenger cars that will only move human beings, trucks will carry different types of goods. It can be difficult to fully fix some goods such as protruding rods, and these goods can kill people and damage the truck during a hard braking.
- Trucks are much heavier. A hard braking will cause fires due to the heat generated. Since most trucks will be very heavy, and there will be a large friction between the tires and the ground surface. A hard braking will generate a lot of heat, which will in turn cause fires.
- Trucks are much longer. The weight will shift forward front wheels during a hard braking, making it difficult to control the trucks. Since the distance between rows of wheels is longer for trucks than for cars, the weight of trucks will not be evenly distributed on all wheels while shift towards the front wheels during a hard braking. This shift means the contact area between a truck and the ground will be smaller, and the friction will be smaller accordingly. A smaller friction during a hard braking means it is more difficult to control/stop the truck.
- Trucks have more wheels, and this can cause more malfunctions of the ABS system. Since trucks have more wheels, it is more possible that the ABS will be activated to limit the brake once there is one wheel/sensor not working well.
- Trucks generally need stronger braking force to get fully stopped, and few systems can stop a high-speed truck quickly.
3. Different technical solutions
Due to the aforementioned reasons, it is reasonable that there will be different autonomous driving solutions for trucks and passenger cars. To be more specific, the differences in solutions include:
- Map: Since most trucks will be operated on highways, the routes will be relatively fixed. This means HD map is not a big issue to autonomous truck companies. For autonomous car developers, they have to seriously consider the high cost/difficulty of gathering and updating HD map data for large areas, since no one can predict where the users want to go. While for autonomous truck companies, they can easily maintain the data for the several routes that the autonomous trucks will be operated on.
- Perception: There is a huge difference in perception solutions for cars and trucks. Since trucks cannot do hard braking frequently and need a much longer stopping distance, trucks need a long-distance perception solution. Based on TuSimple’s experience, 300-meter is the minimum requirement for operating trucks safely on highways. Unfortunately, LiDAR only has a valid detection range of 80-120m, while cameras can see over 1,000 meters. Therefore, autonomous trucks will primarily rely on cameras instead of LiDARs. Therefore, having a good long-distance perception solution can be a key to success. Moreover, a combination of cameras and LiDARs will be the only perception solution for autonomous trucks, while autonomous cars can potentially only rely on LiDARs.
- Localization: Trucks are much longer than passenger cars. You will need to locate both the “head” and the “body” of a truck before taking any actions. For example, when in a narrow space, it is possible that the “head” can turn around while the long truck body cannot. However, this is not an issue for cars.
- Planning: Since trucks and cars are operating in different environment, it is quite natural that they have different planning solutions. Moreover, the priority of these solutions will also be different. For autonomous trucks, they need to be: safe, highly efficient, and cost-saving. Therefore, for example, it is better to cut the lane than to brake (to keep safe and save tire). However, for the autonomous cars, it might be better to brake than cutting the lane (in the urban areas).
- Control: The electrification of trucks is much slower than that of passenger cars. Most autonomous truck companies are developing their solutions for internal-combustion-engine-based vehicles (ICEVs), while most autonomous car companies are developing the solutions for electric vehicles (EVs). The control system for these two types of systems are very different, and therefore the “control” part of the autonomous driving system will also be very different accordingly.
Therefore, it is not easy to transfer the autonomous driving system from passenger cars to trucks.
Strategy: autonomous driving solution provider or service provider?
Eventually, the independent autonomous driving system developers will be acquired or merged by either manufacturers (OEMs) or service operators (shippers/logistics companies/mobility service companies). However, it can be a long time before the final integration will happen, and there is a precondition for the integration to happen: the autonomous driving market has matured, and the technology is widely available.
At this stage, no one can predict how long it would take before the technology matures and even becomes a commodity. However, these independent autonomous driving companies may need to think about the pros/cons of different strategies.
Building a large fleet and providing AV-based logistics service can build a high market barrier, while it also requires a high upfront investment, which can limit the flexibility of start-ups. On the flip side, having no plan for fleet operation can also be risky, since more and more companies are getting into this space, particularly the autonomous truck space, and it seems like the technologies from different companies are quite similar.
Investment: how to compare different autonomous driving companies?
Other than the common things (e.g. team, technical solutions) that most VCs will look into, the following factors can help when evaluating companies.
- Product development progress: Many autonomous driving companies will offer “Demos” when attracting investors. However, these demos can be of little value. The order of values of different information is as follows:
Regular commercial operation (including trials) > Early-stage commercial operation (including trials) > Investors’ ride on public roads > Investors’ ride in closed areas > On-site demo (no ride offered) > Unedited demo video > Edited demo video (little value) > Media report (little value) > Company presentation slides (little value)
- Partnership: For independent autonomous driving system developers, it is crucial to build up an ecosystem that will supply the hardware (e.g. chips, cameras), the vehicles, and related services (e.g. bodybuilders). This is a point that a company can build its competency.
- Productization: The product is different from the technology. In order to turn the technology into a product, a company needs to do tons of market research to identify hundreds of metrics of the product. For example, a company needs to define what types of trailers will be used in the truck, under what types of weathers will the truck be/not be operated, and what is the most favorable operation range, etc.The productization of a technology is extremely important, since it will guide the development team to design something that the customers really need. Almost every technical challenge can be resolved in the end, but not every solution can be a product to be sold on the market.Eventually, in the autonomous driving industry, companies are not competing by solving the most complicated problems but by being the first to go to the market.
- Commercialization: Since many companies have already signed contracts with large shippers and logistics companies, it is reasonable to consider the commercialization progress when evaluating an autonomous driving company. If a company has been providing regular commercial service and generating recurring revenue, then it is a great news. If a company has not acquired any customer (even for trial) yet, there might be a long way to go for this company.