So how AI chips and algorithm has changed the autonomous driving in many ways?

Mr. Shi: "I believe finally the massive deployment of AI in autonomous driving would be edge computing, which means all AI chips are deployed in the vehicles. So we also need to reduce the power comsumption of AI chips as we increase the computing power. In this way, I think the AI chip complexity should be simplified, while providing more computing power at lower power consumption for the massive applications in autonomous driving.”

Mr. Xia: "We need more powerful chips in order to process the data that are generated by all the sensors. Therefore you need those chips in the car, not the cloud. Cloud computing takes 100s of milliseconds to process the data and send it back to the car, which is bad. Therefore we need an embedded system in the car to make the computing process instant. In the future, V2X infrastructure will become very important for autonomous driving. Roadside cameras, lights, etc. will be fed to your car which can help make decision making in terms of speed."

Mr. Sebastian strengthened the importance of hardware; the majority has to happen in the hardware. “Vehicles in the past were still Windows 95 under the hood, and were not designed for mass computation. What they wanted to do at Evoke was take the internal computations of the car and put it into a bike. Therefore, the chipsets were not designed for the motorcycle conditions they wanted to put them in."

In terms of the hardware, what are the pros and cons of lidar and camera?

Mr. Shi believed that ultimately LiDar will be the future, especially for a higher level of autonomous driving: "Two campaigns are ongoing right now (Waymo vs. Tesla). They make their decisions based on the level of their autonomous solutions. Waymo needs to ensure 100% safety for passengers while Tesla takes less responsibility. 

Mr. Xia thought that startups may not be able to become big players in autonomous driving as OEMs but can become OEMs’ suppliers: "we need capable hardware before we can have a sound autonomous driving system. It could take longer to see the commercialization of autonomous driving and is still very premature. It might still need 5 years to be able to see commercialization.”  

Mr. Sebastian was not going to speculate for LiDAR or Camera: "They do very basic sensor technology. Autonomous driving, as a motorcycle rider, is scary because of how they are programmed and we need to be more concerned about 'what if' situations, like the acceleration rate of the motorcycle vs the autonomous software. Therefore, the human factor is the hardest to understand for these data-driven tech companies."

Trucking or Robotaxi, which one will be the first form of automotive autonomy to take off commercially?

Trucking and passenger Robotaxi are the most popular landing scenes for autonomous vehicles now – e.g. Waymo, the self-driving unit of Alphabet, which is focusing on ride-hailing, and TuSimple (图森未来) which focuses on developing Level-4 (SAE-standard) commercial autonomous driving truck solutions. The question of which will arrive earlier has been widely discussed.

Mr. Xia thought that Robotaxi is the better choice and more feasible within cities because trucks face very different conditions on both highways and cities in China.

Mr.  Shi believed other scenarios will commercialize even faster, such as delivery, road cleaning and trucks in harbors – those applications may ramp up in 2 years.

Mr. Sebastian thought it will depend on the environment; the first arrival will happen in well-prepared industries with demand behind them. 

Speed up the process

Through 2018, we have seen a huge acceleration in investment in autonomous vehicle technology. According to KPMG, Europe and North America are more ready for this new mobility. The regional difference is huge, which shows the importance of testing, validation and homologation.

The arrival of fully autonomous cars might be some years in the future, but companies are already making huge bets on what the ultimate archetype will look like. To speed up, we need more substantial progress on fundamental fronts – in the legal framework, infrastructure, investment and consumer acceptance, which will speed up the arrival of autonomous technologies.

Another way to speed up the process, according to Mckinsey, would be to make the shift to integrated system development. Instead of the current overwhelming focus on components with specific uses, the industry needs to pay more attention to developing actual systems, especially given the huge safety issues surrounding autonomous vehicles.