Mr. Liu Xuan, VP and Technology Partner
Liu holds a Bachelor of Science degree from the first Yao Class at Tsinghua University and a PhD in Big Data from the National University of Singapore, specializing in data fusion and big data systems.
He was a former Senior Engineer at Google. He has a wealth of experience in autonomous driving, and was Baidu USA’s self-driving unit core engineer and a main contributor of Baidu Apollo project.
DVN : Can you tell us a little about how Deeproute.ai was founded and what the key areas of focus of the company are?
Deeproute.ai was founded in 2019 by Dr. Maxwell Zhou. We are an AI company, Headquartered in Shenzhen. Our goal is to develop a model for Road Artificial General Intelligence (AGI) , solve AI problems in the physical world and lead the AI industry. The current focus is on smart driving.
DVN: Deeproute uses VLA and end-end AI solutions in its latest software stack. Can you talk about the advantages of VLA and end-end ai versus previous perception and separate planning/control methodologies ?
There are now three generations of Autonomous Driving AI techniques. Early systems used modular (perception, planning, control) rule based solutions. The second generation end-end architectures which we see today integrated this into a single neural network which was an improvement as there is no information loss between modules, and it was better at driving for complex (urban) scenarios. The latest techniques use VLA models – Vision Language Action, which enhanced the planning with reasoning capability. Logical reasoning makes for better defensive driving, and the reasoning means the model is not black box, so it is easier to debug and update and performs better for Irregular obstacle recognition and has memory based control.
DVN: You have deployed software in over 100,000 passenger cars – is this primarily L2+ systems and what OEMs are using your software ?
Yes, our current solution is primarily for L2+ systems and includes advanced urban NOA. It supports OTA and improves as more data is collected through driving. Great Wall Motors and smart are shipping the system and there are other OEMs using it that have not been publicly announced. Systems using our new VLA architecture are coming to market later this year.
DVN: What about plans for the robotaxi market ?
Our approach to robotaxi is to do this with mass produced vehicles. Our system does not use HD maps which significantly reduce costs. We are talking to OEMs and local governments about deploying such systems, but the exact rollout strategy has not yet been decided.
DVN: Do you use HD maps and how is the map data sourced/created ?
We have no plans to use HD maps and our end-end and VLA architecture does not reply on this. The cost is high to do the mapping and it needs frequent updates. There is also limited coverage in China and legislation issues as we expand to different areas and countries.
DVN: You support vision only and vision plus lidar ? In what scenarios does the addition of lidar help ? Is robotaxi in China possible without lidar ?
Lidar can enhance 360 perception. It also improves performance in low-light and other poor weather conditions and is critical in complex urban scenarios. Current robotaxis use lidar and it remains a key safety layer in our opinion. Our system is vendor agnostic, and we have worked with Hesai, Robosense and others.
DVN: You have also done some work in the robotruck market ? How does the software and sensor set differ from robotaxi ?
Used to do truck a couple of years ago but we are not focused on that now.
DVN: What hardware solutions does your model run on ?
Our software runs on multiple hardware architectures. We have worked with Nvidia for a long time. We have also worked with Qualcomm and other China based chipsets.
DVN: What is your licensing strategy with the software ? Have other Robotaxi companies also used it ?
Our focus is on OEM partners and to customize/integrate the solution to different vehicle models.
DVN: What are Deeproute’s international expansion plans ?
The first path to international expansion is with the China domestic automakers as they expand globally. We are also working with China based JV auto makers and talking to customers in Japan, Korea and the EU.