A Chivalrous Pursuit of Innovation: Yiqi Cai's Take on Autonomous Driving
Yiqi Cai is a multifaceted individual. He's not only a partner at DeepRoute.ai but also holds the position of VP of Engineering. He's a meditative thinker, an avid gamer and also a genius who started doctorate studies at the age of 19. He is one of the main pillars in developing mapfree technology and supporting massproduction projects. We had an opportunity to interview Yiqi about his thoughts on technology innovations and the future of AI. His answers offer a sneak peek of the thinking behind strategic shift from Robotaxi operations to L2+ mass production.
About Technology and Mass Production Projects
Q: Why do you think working with OEMs on L2+ smart driving cars is necessary?
A: I believe that technology is not just a pie in the sky. Technology innovation is meaningful only when it offers practical utility and value to people. We aim to build affordable smart driving cars with our solution.
The automotive industry possesses well-rounded technology in manufacturing, integration, supply chain management, and a mature safety management system. So, when we enter the automotive industry, it's important to determine our own positioning.
Q: From L4 Robotaxi demos to L2+ projects, what technology breakthroughs have you made in this process?
A: We've achieved four major breakthroughs.
First, perception has greatly improved, allowing it to perceive road elements that were originally reliant on high-definition maps.
Second, the localization module can closely collaborate with the perception system and determine available lanes based on standard map road topology.
Third, we've fine-tuned algorithms, reducing the planning and control modules' reliance on deterministic map environments to ensure system safety.
Fourth, we've conducted numerous generalization tests under various weather and road
conditions to train the data loop and drive technological iteration. This ensures our system can consistently deliver stable and excellent performance in different situations.
Q: I bet there's a lot of pressure and many difficulties on these projects and I wonder how you and the team cope with it?
A: Indeed, we face a significant amount of pressure.
Back in 2020, when we had just decided to adopt a map-free approach, there were limited map providers available due to the absence of a universal data interface format. We were also not entirely sure how to adapt once the data format was confirmed. However we had solid faith that the map-free approach would be the trend in the industry, and we proactively communicated with the map provider so they could help coordinate the research and development of map-free technology. Eventually, they shared our vision and were very cooperative.
Working with OEMs on a mass production project required extensive communication because smart driving was a novel industry without standardization. Consequently, we internally reshaped our culture to be client-oriented and prioritized delivery.
We had very tight timelines. We also needed to rethink the solution, while at the same time, the product team, testing team, and R&D cooperated more closely than ever before, working around the clock to make things happen.
Q: What are the advantages of data-driven solutions compared tprule-based solutions?
A: There are infinite corner cases in the real-world scenario. If you keep adding rules to the algorithm, it might become unpredictable and lead to inverse optimization. A data-driven intelligent driving system, on the other hand, specifies a model's cost function, establishes a closed-loop process for data feedback and accumulation, and delegates the optimization
process to the machine, replacing human decision-making. Therefore, the data-driven approach ensures better system optimization while enabling collaborative development efforts across a larger group of people instead of relying on singular decision-making.
Q: Can you share your sense of accomplishment when working with the OEMs on mass production projects?
A: There are quite a few. When we first received the offer from the OEM, I felt proud and excited that our technology had crystallized and could bring value. Meeting each deadline to deliver the required functions filled us with pride, even though there was some pressure. When we were on the same road, competing with rivals' products, I discovered that our product offered a smoother lane change experience, better riding experience, and safer maneuvers. That is the closed-loop process for my sense of accomplishment.
About AI and the future
Q: Will human drivers be replaced by machines?
A: Definitely. Whether it's from first principles or other angles, you would realize that human driving is not reliable or consistent because humans generally have limited attention spans, inadequate observation distances, and are susceptible to mood swings. Machines outcompete humans in all of these aspects.
Q: So when do you think autonomous driving will become a reality?
A: It still takes some time. But I believe it will evolve exponentially. It has been only 7 years of development since autonomous driving officially takes off in 2016. It's only in recent years that we've witnessed flourishing progress happening in AI development, computing power, and people's acceptance to smart driving, among other factors.
About Personal Life
Q: We also heard you're very into martial arts and chivalry novels. Can you make an analogy between mass production projects and one of the novels?
A: This is a tough question. Given the landscape of autonomous driving industry, I think it's quite like the novel "Demi-Gods and Semi-Devils" by Jin Yong, because there are different schools and disciplines of martial arts and each of them is unique. It's difficult to say whose martial arts is better than others.
Q: From your perspective, in the novel "Demi-Gods and Semi-Devils", whose role does DeepRoute.ai play?
A: We are very strong in perception. In the context of autonomous driving, perception is similar to the inner power known as Neigong in kung fu. It lays a solid foundation for all technological development. In this context, I think if DeepRoute were one of the characters in this novel, I hope it would be the hero, Qiao Feng, who was a top-notch martial artist known for his discipline, humility, and mastery of distinctive and impactful techniques.