Autonomous driving - are we there yet? (Part 2)
In the second part of this user opinion series, written by Jack Creasey, he'll guide us through the development of hardware and software capability for autonomous cars.
What about the development of hardware and software capability for driverless vehicles?
I think that it’s difficult to discuss vehicle automation without also discussing SAE levels of capability for vehicle automation. SAE have defined Levels 0 – 5 in the standard Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles J3016. It is important to note that any vehicle may have multiple SAE Levels that apply to it at various times. For example a Tesla Model S has an SAE Level 2 capability; Autopilot. If the user does not turn on Autopilot, then the Tesla has mainly ADAS components that are part of the SAE Level 1 definition (Wikipedia has a great ADAS list). I won’t talk about all of the SAE Levels in this article. For consideration, let’s define in very general terms just SAE Level 2, 3 and 4.
SAE Level 2: A human is in control and responsible for the vehicle at all times. Even though at this level the vehicle may follow lanes and on/exit ramps, automatically brake and accelerate and manage stop and go traffic, the system may at any time and with very short notice (a little as a few seconds) require the human to take control. In production vehicles this level of capability may be a collection of automation features where the driver can make choices about which features to enable.
SAE Level 3: A human driver is required to be available on request, but the automation will only request human intervention with a reasonable changeover time. Much like SAE Level 4, SAE Level 3 is likely to be geo-constrained but within that area does not require the driver to be attentive at all times. This level of automation is not suitable for a driverless vehicle since it may call on the driver to take control for some situations.
SAE Level 4: The automation at this level is able to handle the complete driving task within a defined geo-fenced area. It will never need to call on a human to intervene within the geo-defined area and can always take the vehicle to a safe state no matter what problems arise.
Let’s use three of the potentially top players as a comparison of where we are today with the SAE Levels:
Tesla is currently producing its own vehicles (playing both the automation developer and vehicle manufacturing role) with SAE Level 2 Autopilot on both Level 1 and 2 Hardware (HW) with an announced future capability that enables driverless operation. Tesla’s Autopilot capability varies between a Level 1 and 2 HW, but they are already advertising that an upgrade to Level 3 HW will be required to reach full driverless capability. Tesla is a great example of incremental development, but in my opinion, they do not have a great reputation for delivering product. However, the option is there to incrementally progress through SAE Level 2, then to Level 3 (still with a human as a form of backup) and eventually to Level 4 in constrained areas with no human backup.
When might Tesla reach fully driverless (SAE Level 4) capability? That is really an unknown, but I think that it is likely to apply only to new production vehicles and unlikely before 2020.
Secondly, Waymo have a target of SAE Level 4 for their automation controller, and their advertising claims that they are building a driver not a vehicle (which is an intriguing definition). Waymo has no vehicle production capability (even Firefly was outsourced), and it must partner with multiple vehicle manufacturers to produce a driverless vehicle end product. In my opinion, this deficit will present a sizable challenge since Waymo must build a bridge into the vehicle controllers (ECUs), and for each manufacturers range of vehicles, this will be different. As Waymo’s fleet size and vehicle types expand to meet future demand over several model years from any given manufacturer, vehicle maintenance and revisions are likely to become a major challenge.
In my opinion, to provide a viable driverless vehicle Waymo must solve all the problems, that all their vehicles might be exposed to in a given geo-fenced area. While their electronic driver (AI) provides learning for the whole fleet, as the complexities of multiple cities are tackled, the number of problems they must solve increases exponentially. To date their largest deployment is in Chandler, AZ, and here they cover just 100 square miles. Waymo have announced that they are now providing a public service in Chandler, but they have downgraded to using safety drivers in all vehicles instead of fully driverless service. Waymo also announced the use of remote backup drivers to handle unexpected events.
Thirdly, GM/Cruise has a target of SAE Level 4 for their vehicles, with no human backup support in the vehicle. Undoubtedly, GM has a stable of platforms available with the Chevy Volt being the development choice so far. Like Waymo, Cruise has to solve all the problems in a specified geo-fenced area to achieve a production vehicle or introduce remote backup drivers which in my opinion would reduce them to SAE Level 3 capability.
The definite advantage with the GM/Cruise combination is that they are vertically integrated to a large extent and can do everything within their own production facilities.