In the race toward self-driving vehicles, developing novel hardware or software is not enough for newcomers challenging major automotive suppliers and technology companies. They must also secure funding and get permission to test vehicles on public roads to gather valuable data. DeepRoute.ai LLC managed to accomplish these tasks last month.
The California Department of Motor Vehicles (DMV) issued DeepRoute an Autonomous Vehicle Testing (AVT) permit. It allows the company to test its self-driving cars with safety operators. There are 63 other companies with AVT permits in California.
DeepRoute said it plans to begin testing its Level 4 autonomous driving system (ADS) around Silicon Valley with the goal of perfecting its Early Fusion sensor-fusion technology. It also intends to refine its path-planning algorithm that works with its simulator and data platform.
The company, which has offices in Shenzhen, China, and Fremont, Calif., also raised $50 million in funding last month. DeepRoute was founded by a team with experience from Google, Microsoft, and Ford. It said it plans to hire additional staffers and expand globally by year’s end.
Shuang Gao, chief operating officer at DeepRoute, described his company’s approach to The Robot Report:
Can you briefly describe DeepRoute’s sensor-fusion and planning technologies? How are they different from those of competitors?
Shuang: Our sensor fusion is based on deep learning technology, which combines multiple sensors including lidar, cameras, and radar in high-level features to detect the obstacles all around the self-driving vehicle.
For planning, we use an end-to-end neural network to directly output 3D objects and their future trajectories and finally select the best planner path for self-driving. Other self-driving companies may be divided into stage-by-stage subtasks, like perception, tracking, prediction, planning, and control.
What has been the biggest challenge for DeepRoute so far — developing technologies, finding partners, or negotiating regulatory requirements?
Shuang: In our industry, the most challenging aspects tend to be a combination of technology development and regulatory requirements. As we are entering new territory with autonomous vehicles, it is imperative that safety comes before everything else. Our team is working hard to ensure that each technological advance we make ensures safety and [complies with] regulatory requirements [while also pushing] the boundaries [toward] further progress to full autonomy.
The technology is the foundation of everything we do at DeepRoute. If our technology is advanced enough, the rest will just follow. The core of the L4 self-driving stack is the “perfect perception” and ultimate safety. To achieve it, we developed an L4 full-stack self-driving system including modules such as sensor solution, sensing, planning and control, HD map, etc. The whole system has been completely designed and developed by our team to a high safety standard.
How will your products lead to safer autonomous driving?
Shuang: Our mission is to build the safest, smartest, and most reliable autonomous driving technology. We uphold the highest safety standards and consider all operators, passengers, and pedestrians.
We cooperated with leading OEMs and produced self-driving vehicles following the automotive-grade [standards]. DeepRoute follows industry guidebooks including ISO26262, SAE3061 and GB/T34590. This would ensure the safety of self-driving vehicles, electronic components and networks, and mass production.
How long did you work on them before applying for an autonomous vehicle testing permit?
Shuang: We’ve been working on this continuously since the day we were founded in February 2019, and we are proud that we can test our vehicles on the public roads in California. It is still an ongoing process, but I can tell that our advanced technology leads to safer road testing.
Are you working with any automakers or other technology partners? Whose sensors are you using?
Shuang: We are now working with Dongfeng Motor Corp., a leading Chinese OEM. Thanks to our self-developed ADS controller, we can easily adapt to and precisely coordinate different sensors from different suppliers. The major sensors are lidars, GNSS [Global Navigation Satellite System receivers], and cameras from multiple suppliers to keep making the ADS better and better.
How difficult was the process of applying for an AVT permit from the California DMV?
Shuang: It wasn’t too difficult, but it took us a little over a month to get an AVT permit from the DMV. We had to make sure that the safest drivers are qualified under DMV regulations and requirements.
We had to enroll in the Employer Pull Notice [EPN] program as well as insurance and filed at the California DMV to get them reviewed and qualified for an AVT permit. Finally, we are testing our vehicles on the public roads in Silicon Valley and excited about the results.
How are you going about finding and training safety drivers?
Shuang: Currently, we have several DMV-qualified safety drivers. We’ve found drivers who have previous experience with autonomous vehicles. Our training program includes defensive driving, engaging and disengaging the autonomous system, data-collecting procedures, and more.
What have you accomplished with your manufacturing partner?
Shuang: Dongfeng is one of the Chinese leading OEMs, and it worked with DeepRoute to support vehicle platooning technology and provide robo-taxi services at the 7th CISM [International Military Sports Council] Military World Games.
DeepRoute ia planning robo-taxi service for Japan. Why there? How are rules and expectations different?
Shuang: Japan was chosen because of the great potential of the Japanese autopilot market. The country’s transportation system is very standardized, and local people’s compliance with traffic regulations is strong. The Japanese population is aging and has a strong invisible demand for autonomous driving. The country has a good business environment, and its automotive industry is very developed.
In terms of the policy, whether in high-precision mapping, technical testing, or the promotion of experimental projects, the development of the Japanese autonomous driving industry has the participation of the state, enterprises and private institutions.
The Japanese government is now able to achieve Level 2 autonomous driving at high speeds, and by 2020, the law will allow Level 3 vehicles to drive automatically. The commercialization of L4-class autonomous driving in a specific area is expected to be achieved by 2020.
We are still looking into global markets for our robo-taxis. For any market that we are operating in, we make sure to comply with the local rules and go beyond expectations.