Tesla, Waymo, Cruise and others are all frantically developing self-driving technology. Will anyone actually do it?
Episode Summary:
Chinese technology firm Baidu, through their Apollo AI unit, is looking to take the lead in the race to the robotaxi. Will they win? While leadership in the newly emerging field is hotly contested, there is a possibility that few have considered: that no one may win.
Access all episodes of End of the Line on engineering.com TV along with all of our other series.
Transcript of this week’s show:
We talk a lot about self-driving cars here at engineering.com, and for good reason. It’s the hottest technology of the decade, for one, and it’s an area where multiple players are racing for the lead, or more accurately racing to be first to deliver an actual self-driving solution. The Chinese company Baidu Apollo has thrown their hat in the ring, launching level for Robo taxi system in a Beijing suburb. They join Cruise in San Francisco, and Waymo in Phoenix as service providers of truly driverless taxis, although under controlled conditions. No matter what the system, they all operate under restrictive conditions, whether it be visibility, or the spatial limitations of their geo-fenced and carefully mapped operating zones. Tesla is going after a universal solution using machine vision to eliminate the need for extensive mapping and geo-fencing, but their autopilot product has been delayed for years, has had at least one foundational rewrite and is still in beta. Five years ago, industry experts predicted that the problem would have been solved by now, but some experts feel we’re still a decade away. Why? I’ve already commented on potential regulatory issues, but in technological terms, there has traditionally been two schools of thought: one is to give the taxi an inbuilt map of its environment, so it already knows where it is and infills local, transitory information through sensors, like people and traffic. The other is to use onboard AI to take real-time sensor data and make decisions the way human drivers do. The geo-been solution reduces demands on algorithms. Knowing where you are at a moment in time is a big part of solving any driving problem. The Tesla, machine vision approach emulates what humans do, the neuroscientists are typically sure how we do what we do behind the wheel, and it’s computational load is significantly heavier. The Baidu Apollo approach is interesting. While their initial system looks a lot like the cruise concept, Apollo is working on citywide traffic management systems that link thousands of vehicles to central controller through the cloud. For a Robo taxi, an interesting way to lighten the computational load and reduce the chance of an accident, by controlling the other cars in the intersection or on the roadway. With every car avoiding every other car, edge cases should be less likely, and overall system safety should improve. I can see this working in China, but have difficulty imagining how a Western city could mandate total central control of every vehicle, even in a congested downtown core. So, does this give the Chinese company the advantage? We don’t know but barring some breakthrough in processor technology or AI algorithms, I think we’re a decade at least from SAE level V self-driving. And there’s a very real possibility, that it may not reach human levels of skill in our lifetime. But I hope I’m wrong.
