New intelligent driving: Drive.ai has performed very well in the past two years in terms of financing, team, and technology implementation. Especially since the second half of 2017, its actions have been updated frequently, and it has won the favor of well-known venture capitalists such as NEA and GGV, and has received a $15 million investment from Grab, the largest ride-hailing company in Southeast Asia, and launched a trial operation project with Lyft, the second largest travel platform in the United States. The Stanford team is rapidly expanding its "circle of friends."
Drive.ai aims for Level 4 autonomous driving technology, and its practice route is dominated by deep learning, and it has also undergone a transformation from vision-first to sensor fusion. In October last year, when the company met with co-founder Wang Tao in Silicon Valley, the latter also revealed that the team was preparing for a larger pilot project outside of the Bay Area and Singapore. At the moment, the company has not disclosed its progress.
On January 16, 2018, Lei Feng Wangxin Intelligent Driving will hold the GAIR Silicon Valley Intelligent Driving Summit in Silicon Valley, USA. At that time Drive.ai co-founder Wang Tao will also be present to share the rich technology and the latest developments of the Drive.ai. Click "Read More" at the end of the article or visit
You can learn more about the GAIR Silicon Valley Intelligent Driving Summit and participate in this luxury event of autonomous driving in China and the United States.
Following a $50 million Series B funding round in June 2017 led by NEA and followed by GGV, Drive.ai and Lyft announced in early September that they would pilot a robotaxi service in the San Francisco Bay Area. In the same month, Drive.ai announced a $15 million investment from Southeast Asian taxi giant Grab, which will set up an office in Singapore in the coming months.
California and Singapore are arguably the two regions with the most open and well-established autonomous driving testing regulations in the world. Drive.ai a robotaxi service in partnership with Lyft is about to go live, and Singapore will be the first stop for its expansion in Asia. At this time, the reporter came to the Drive.ai headquarters in Mountain View.
The third year of entrepreneurship
In 2015, six PhD/Master students from the Stanford Artificial Intelligence Lab took a collective sabbatical and invited roboticist Carol Reiley and senior business consultant Fred Rosenzweig to join a large founding team. By April of that year, the team had received $50,000 in seed funding and started a real garage venture at the Silicon Valley Innovation Incubator in Santa Clara, Silicon Valley.
Wang Tao, co-founder of Drive.ai, told us that there are now more than 80 people in the team, and more than half of them are making software. Because Drive.ai adopts a deep learning-first strategy in the design of the entire autonomous driving scheme, in addition to conventional perception, positioning, planning, and control, a lot of software work is related to deep learning:
By 2017, Drive.ai had completed the adaptation of autonomous driving technology for three different models, including the original Lincoln MKZ, later the Audi A4, and more recently, a van. Wang Tao told us that this means that Drive.ai's autonomous driving system can be transplanted between hybrid, gasoline, commercial and other models.
The addition of new vans also means that their technology can be used not only to carry passengers, but also to transport goods.
Technical route
Drive.ai's technical route has always emphasized deep learning first.
From the perspective of sensor solutions alone, their retrofit kits for the aftermarket have gradually taken shape. In the case of the Audi A4, for example, they connected four 16-line lidars directly above the roof, placed two 16-line lidars diagonally on either side of the roof, and used multiple fisheye cameras on the top and sides of the roof, adding a millimeter-wave radar to the front of the car. The sensor scheme used by different models is slightly different, and the team will adjust some of the technology options as the sensors available on the market change. For example, the new van uses LiDAR with a higher wiring beam, and the number of LiDARs used in the whole scheme will be reduced accordingly. The choice to concentrate most of the sensors on the roof also made it easier for Drive.ai to deploy the solution in bulk.
On the Drive.ai's actual car, we can also see an LED screen under the four lidars to display information about interaction with other vehicles and pedestrians. There are also processors integrated behind the screen that pre-process the data acquired by the sensors.
In general, a Drive.ai retrofit kit will consist of a sensor solution, a screen for interaction with the outside world, and a packaged processor. The three pieces of hardware are integrated to form a module on the roof of the car, which is then connected to a computer in the trunk. Wang Tao said that at present, through algorithm optimization, the power consumption of the entire system is about a few hundred watts, which is equivalent to a high-performance gaming PC. A gasoline vehicle like the Audi A4 can also support the system.
In the garage, Drive.ai has prepared nearly 10 prototypes. At present, the mode of external cooperation is mainly as follows: OEMs open up some vehicle CAN interfaces, Drive.ai complete drive-by-wire transformation, and then connect the software and hardware of autonomous driving. The first vehicles will begin operations through commercial fleet operators such as Lyft or logistics fleets.
Drive.ai and Lyft are about to launch a joint taxi service
On Sept. 7, Drive.ai and Lyft jointly announced that they will soon offer a pilot project of robotaxi service to the public in the San Francisco Bay Area. In this project, Drive.ai will form a mixed fleet using the current three models. According to Lyft's description, passengers are likely to ride in a self-driving car if the route they choose happens to be in the area covered by their high-precision map. Wang Tao told us that Drive.ai and Lyft will be working on a new app specifically for this purpose. Unlike the existing Lyft app, the app will leverage Lyft's vehicle and passenger dispatch backend and incorporate Drive.ai design at the app interaction level.
That's because when the California DMV issued a self-driving road test license, the self-driving companies had already paid a $5 million deposit and purchased insurance for the test vehicle. As a result, in terms of technical frameworks and regulatory processes, robotaxi services are ready to be made available to the public. At the moment, Lyft and Drive.ai are more discussing the details of the technical implementation.
Wang Tao revealed to us that in addition to the two major cooperation with Lyft and Grab, there will be larger-scale landing projects in Drive.ai.
The following is part of the dialogue between Wang Tao and Xinzhijia, and Xinzhijia has edited the content without affecting the original meaning.
New Intelligent Driving: Your in-vehicle information communication system does not use ROS, is it because ROS is too inefficient?
Wang Tao: I know that many companies in the industry are using ROS, but because ROS is not designed for autonomous driving, this system is designed by ourselves.
ROS is a product of academia, and it was designed to run on robots. It has several advantages, one of which is open source. After it is open sourced, everyone will use it, and there will be a community, but this also creates a lot of redundant tools. A lot of things are not designed for autonomous driving, and a lot of things can cause unnecessary operations.
After verification, we felt that ROS was too redundant, not streamlined enough, and not focused enough, so we decided to make a better system ourselves, which is more suitable for the needs of autonomous driving.
New Intelligent Driving: How are Drive.ai currently working with car companies that use your products? Can you give an example?
Wang Tao: Our previous cooperation model was mainly for car companies to open CAN interfaces and wire control interfaces on the car, but the wire control is realized by ourselves, and the bottom interface still needs to cooperate with car companies.
At the moment our main partner is the commercial fleet, which is the first step in helping them to automate and install turnkey solutions on the fleet to help them save costs.
I think these commercial fleets will be the first to apply automation and autonomous driving technologies on a large scale, because they are not as price-sensitive as the end consumer and are not as strict about the appearance of the car.
At present, to create a true L4 or L5 autonomous driving system that does not require a driver, many redundant sensors are needed. If we really want to hide all the sensors outside (the whole system is up to mass production standards), even if a large car company wants to do this, we think it will take a few years to complete the integration work. In the short term, the car will definitely look more sci-fi, and end consumers may be less receptive to this look.
But for the team, they don't really care about how the car looks. It looks a little different and may even be beneficial to their own propaganda.
We can mainly help them save costs. Because in the U.S., the cost of the driver currently accounts for about 70% of the cost of the fleet, so this part is relatively large. If we can help them save some of their costs, it will be a great help to them.
New smart driving: Drive.ai work with partners, such as deploying retrofit kits with fleets. What are the objectives of OEMs in this regard? What other ideas are there besides open interfaces?
Wang Tao: At present, the main business model of OEMs is to sell cars to end consumers, and most of their profits come from this field.
Now there is a trend of autonomous driving, and they will follow this trend, and will make some attempts, mass produce some models with L2 and L3 functions, and even aim at L4 cars, and they may also form some fleets themselves.
I know of a few automakers that are deploying their shared mobility fleets, and this is their next possible growth point. But if everyone chooses to share the car, then no one will buy a car, which will also have a certain impact on their own business model.
One of the strengths of car manufacturers is system integration, which integrates sensors, lidars, cameras, millimeter-wave radars, computing platforms, etc. into the car, which is the strength of car manufacturers. Once a solution is in place, it is possible to reduce costs and start mass production, which is also a strength of car manufacturers.
But in this process, I think they also have a lot to learn, for example, they have a set of hardware, but generally speaking, the industry does not think that the car factory has too many advantages in terms of software, and we feel that we can give the car factory some help in this regard, which is also one of our cooperation models.
New Intelligent Driving: Will you cooperate in the field of L3, because car companies are still very interested in the mass production of L3?
Wang Tao: L3 is not our focus at the moment, the company's focus is still on L4.
New Intelligent Driving: In terms of operations, Drive.ai will cooperate with the fleet. What is this convoy? Is it a platform like Lyft, or a more traditional trucking fleet, or a commercial vehicle fleet?
Wang Tao: In terms of operation, we hope to become an autonomous driving platform, the core autonomous driving algorithm is the same, and we can open the service interface to these fleet operations, for example, if a logistics company wants to cooperate with us, we can also open the interface to them.
But the core technology of autonomous driving is still in our hands.
New Smart Driving: Drive.ai Three car models have been deployed. What is the difference between them to you? How did you choose these three models?
Wang Tao: At the beginning, if you want to build a system quickly, the Lincoln MKZ is a good choice, and the Audi A4 is also our first attempt to deliver a turnkey solution that can be installed on different models without major changes.
For the third commercial vehicle, we also demonstrated our ability to perform drive-by-wire on commercial vehicles, while our system does not require major changes in commercial vehicles.
New Smart Driving: Drive.ai is working with Lyft, will you put these driverless cars on the Lyft platform?
Wang Tao: There is no need to use Lyft's platform completely, there may be some differences, because the user experience is still somewhat different.
We also wanted to build our own Drive.ai platform, which would be different from the normal Lyft in terms of user interface, and be a joint app.
New Intelligent Driving: We just talked about Drive.ai are also developing their own high-precision maps. What the hell is this?
Wang Tao: We are still using high-precision maps (developed by ourselves) internally, but the degree of dependence is not so high, and the way of dependence is more flexible.
For example, when some companies use high-precision maps, they determine the location of lane lines through the reflection value of lane lines, and match them with the map for positioning.
But we found that once it rains, the reflectivity of the lane lines changes greatly. That's why some technical solutions may not be used when it rains heavily, because a thin curtain of water will form on the ground, so the radar light will rarely return after hitting the ground. This is actually a difficult point.
We take a slightly different approach to positioning, so we don't have to be too sensitive to the environment. We have published a video of the road test in the rain.
New intelligent driving: Drive.ai said before that it can not only carry passengers, but also transport goods, is there any difference between the two?
Wang Tao: There is not much difference for us, because when we choose the model, we also choose a platform that can carry people and goods.
The goods will be transported through cooperation with logistics companies, and negotiations are currently underway.
New Intelligent Driving: After the completion of Series B financing, Drive.ai has plans for retrofit kits, how is it progressing now?
Wang Tao: Retrofit kit is another name for turn key solution, Retrofit is the meaning of aftermarket, we are not pursuing the perfect combination with the car now, we will transform the existing vehicle, ensure the iteration speed, and implement the autonomous driving technology earlier.
New Intelligent Driving: Will You Consider Making Your Own Chip or Customizing the Chip?
Wang Tao: Not for the time being.
New Intelligent Driving: What are your goals for 2017?
Wang Tao: In 2017 we will have a larger project, which has not yet been announced.
New Smart Driving: Waymo wanted to redesign the car in the early days, but didn't end up redesigning the car, and ZOOX is now redesigning the new car.
Will there be a big difference between OEMs and autonomous driving companies when it comes to the exterior design of future cars? What do you think the car of the future will look like?
Wang Tao: I still say that building a car is not an easy thing, although you may think that the automotive industry is not a high-tech industry, but it is a very, very mature industry, developed for more than 100 years, and the internal knowledge accumulation is not something that a company like ours can do, and it cannot catch up quickly.
Of course, electric cars can be an opportunity to overtake, but I think it will take time. In this process, I believe that autonomous driving cannot wait until electric vehicles are ready to do autonomous driving, and the two must go hand in hand. Because the implementation of some autonomous driving functions will promote the development of electric vehicles.
For example, one of the major problems of electric vehicles is that there are not enough charging stations and the charging time is long. But if you can realize car sharing, you can drive the car to a place, and the car can automatically drive to a nearby charging station to charge. In this way, the threshold for using electric vehicles will be lowered.
At the same time, we know that the internal combustion engine is a relatively complex system, although it is now possible to modify the internal combustion engine for autonomous driving, but compared with electric vehicles, more effort is still needed on the engine model, and the electric vehicle model is much simpler, which is also a benefit for the transformation of autonomous driving. [ENDS]
From CMU to GM, why is Zhang Wende called the "sweeping monk" of the autonomous driving circle? GAIR Silicon Valley Intelligent Driving Summit