What Doug DeMuro got wrong in his Super Cruise vs. Tesla Autopilot Comparison
Doug DeMuro recently posted an opinion piece on his “More Doug DeMuro” channel describing why he thinks GM’s Super Cruise is better than Tesla’s Autopilot. Before reading further, watch Doug’s video if you haven’t already.
As Doug expected, Tesla owners, including me, were quite upset about his conclusion. I’ve been pondering why our reactions are so strong to these takes (“takes” is plural because others have also come to this conclusion).
The surface-level reason why Doug’s take is wrong
In my opinion, GM’s “handsfree” feature (but certainly not attention-free) is simply an insufficient reason to select Super Cruise as being “better” than Tesla’s Autopilot. If you compare the actual capabilities of Tesla’s publicly available Navigate on Autopilot (NOA) features, they far exceed GM’s. I regularly drive with NOA engaged through some amazingly complex scenarios that Super Cruise could never handle. For instance, on my commute, NOA will now regularly drive me home without intervention, including complex merges, interchanges, and wildly divergent traffic flows. As long as I keep at least one hand on the wheel with just a touch of torque applied, the car will do everything else. It will decide when to change lanes and then do so without me lifting a finger. It will detect vehicles approaching from the rear and get out of the passing lane. It decides to merge onto the Interstate based on the traffic around me, sometimes delaying the merge for almost 1/2 mile to find a good time to slot in with traffic by staying in the merge lane that eventually becomes an exit-only lane. It takes my weird exit from the Interstate and then seamlessly transitions down to straight Autopilot through the cloverleaf and onto surface streets. I now regularly find myself dumbfounded by the human-like decisions it makes. I often whisper to myself after it makes one of these moves: “Wait, did it really just do that all by itself?”
On Surface Streets, it now helps ensure I won’t run a red light. It chimes at me if the light turns green, and I don’t press the accelerator immediately. On rural roads and state byways, which I drive a lot, I can use Autopilot even when Navigate-on-Autopilot isn’t available. It vastly increases the safety margin as both human and machine are doing their part. If I get distracted with the radio or a drink, AP is still maintaining the vehicle's safe function.
To the best of my knowledge, Super Cruise doesn’t do any of these things. It lets you have your hands off the wheel. So what? You still have to be paying attention. Even more than with a Tesla on AP or NOA, you must sit there like a robot staring at the road in front of you. In stop-and-go traffic, Tesla’s AP will often let you go minutes without having to touch the wheel, and on long road trips, the ability to look around a bit allows you to arrive at your destination much more refreshed. I will gladly keep one hand lightly on the wheel in return for not having to sit there like a crash-test dummy staring out the front window. If I need to grab some food from the passenger seat while traversing the plains of Kansas, Tesla’s AP gives you more than enough time to do so before nagging you.
The real reason Doug’s take is wrong: The world isn’t a static place.
Super Cruise cruise requires a highly detailed map of the world to operate. If the real-world changes, Super Cruise is dead in the water. It is, in other words, dumb. It doesn’t know how to interpret the world around it beyond the basics of what other vehicles are doing. On the other hand, Tesla’s Autopilot is going through a learning journey to understand the world around it (within the confines of driving). Its ability to sense and understand a dynamic world is already far superior to Super Cruise, and the gap is only growing. In the 16 months I’ve had Hardware 3.0, my vehicle has added the ability to see and react to stop signs, traffic lights, speed limit signs, and has vastly improved its ability to see and understand what pedestrians, bicyclists, and vehicles are doing on surface streets. To be sure, it still makes plenty of mistakes, but Tesla’s underlying premise for Autopilot is that the world is ever-changing and is replete with a gazillion edge cases. And, all of this capability I’m describing is in the publicly released builds. A few YouTubers are in the Early Access program of the Full Self-Driving stack, and Tesla has allowed them to share their experiences. If you haven’t seen those videos, here are three to send you down the rabbit hole of what is coming to all of our vehicles in the not-too-distant future.
Those of us who experience Tesla’s AP getting smarter, smoother, more capable, and able to work in more-and-more situations viscerally understand how GM’s Super Cruise and Tesla’s AP aren’t in the same category, so conclusions like Doug’s strike us (or maybe only me) as missing the mark. On the surface, they appear to both be driver assistants. In reality, one is a driver’s assistant, and one is a neural network that is learning to drive by watching and learning what real drivers do and then integrating that into its neurons…just like humans do. Embedded below are two good videos on the subject. The first is a presentation by Andrej Karpathy, who leads Tesla’s AI team building the autonomy stack. The second is an interview with legendary chip designer Jim Keller (he helped build Tesla’s AI chip and board in every new Tesla) on Lex Fridman’s podcast talking about how Karpathy’s team is rapidly transitioning from computer vision to neural networks and why that matters. The second video's embed link is to Steven Mark Ryan’s take on the interview because it cuts to the chase, but the full episode is available on Lex’s channel.
In Summary
Super Cruise is designed to require a human to oversee its actions both now and in the future. It will never be more than Level 4 autonomous. Even then, it’ll only be L3 or L4 under specific and constrained circumstances. AP is learning from humans so it won’t have to rely on them in the future. Super Cruise is a narrow solution, whereas Autopilot is a generalized driving solution.
Now, as a consumer, you may be happy with a narrow solution. Me? Not so much, so the handsfree feature IMO doesn’t outweigh everything else.
You should subscribe to Doug’s channel because he has great content, but on this conclusion comparing Super Cruise to Autopilot, I think he is completely wrong. What is your view?
