By Chris Tropea
Apr 13, 2023
A personal driving coach may be one of the best options for learning how to go faster on track, but do you need an actual person to teach you?
We take two novice drivers to the track to find out how the artificial intelligence of the Garmin Catalyst competes with a human driving instructor.
Presented by CRC Industries.
12/23/22 3:23 p.m.
One is (mainly) objective, the other is (mainly) subjective. Both are useful.
12/23/22 3:23 p.m.
As I'm beginning to construct the print side of this story, I'm just getting more and more impressed with the Catalyst—and I was already impressed with the Catalyst. The thing it can't do, though, is tell you HOW to go faster in those areas where it's encouraging you to do so. It can give you the information on speed and line, but not technique. But a human coach, using the Catalyst for data and just working with the student on technique issues would have a HUGE advantage when it came to helping drivers develop.
So, as a replacement for a human, the Catalyst is pretty darn good out of the box. As a tool augmenting human instruction, it's a monster.
12/23/22 4:38 p.m.
Great video, and excellent food for thought.
I love my catalyst, but as a (soon to be replaced by a robot?) human instructor of almost a decade, I think you missed the one important "control"in this experiment, which is an in-person instructor. I'm sure you've found the same thing which is that the first few sessions for a never-ever are kinda overwhelming, so the dude that only gets a debrief instruction seems like he's at a huge disadvantage to the guy who doesn't have to keep all the notes in his head in addition to being overwhelmed on track.
I'd also be curious at what rate the improvements came for both.
Either way, excellent content!
12/23/22 5:11 p.m.
As an example of this..
Last weekend I right-seated with an intermediate level driver who was looking to fine-tune his line during one particular session. All I did was feed him cues for where/when to be looking...and his hands/feet did the rest. He nailed the line and speed.
12/23/22 5:51 p.m.
What a fun day. Not very often you get paid to drive a car on track.
On a more serious note, I know there was plenty of time to be found on track. It wasn't until my last few laps that I felt the car reaching it's limits.
12/23/22 6:02 p.m.
We knew this could never be a truly scientific test without the use of either cloning, complete memory erasure or time travel, all of which come with their own logistical and ethical drawbacks. We're simply too busy making magazines and websites to be running around murdering rogue clones, or jumping through the time stream cleaning up preemptive causality violations and bootstrap paradoxes.
So we really just tried to focus on the experiece they were having, and what they might be missing out on by exclusively doing one or the other. As I write the story, I think I'll dive a little deeper into some of the ways that combining these techniques could be extremely beneficial.
12/23/22 6:14 p.m.
is this a "sponsored" post?
12/23/22 6:49 p.m.
In reply to sleepyhead the buffalo :
No. Though Garmin is a partner of ours, this was our idea and Garmin didn't have any input besides providing a Catalyst for the test when we asked. They're seeing this video for the first time today, just like you are.
12/23/22 7:38 p.m.
I believe we are the only people using the dual garmin process. Instructor and student in Identical cars, _+ 2 garmins. We are seeing phenomonal results.
12/23/22 9:10 p.m.
Any level of telemetry is always a good tool but you really want a human coach and or instructor to go through it with the driver.
12/23/22 9:30 p.m.
I really enjoyed using the Catalyst but want to go back to the track again and have JG be able to point out and catch some bad habits I may have started forming.
I am really interested to see in the print story what JG sees in the data from both of us.
12/23/22 9:52 p.m.
Ooh this is interesting. Do the Garmins always agree? Do they ever argue? Is there an Alpha Garmin?
12/23/22 10:55 p.m.
I'm going to re-watch the video, and check out the other content mentioned here, then provide a more substantive comment (hopefully) [in the next couple of days]. I mainly asked this so I could properly calibrate this eventual future comment... not as "criticism" of what y'all put together so far.
12/24/22 9:10 a.m.
My understanding of the Catalyst is that it can only recommend what it has experienced, so it seems like that would be the ideal way to use it as an instructional tool. Or having an instructor drive your car so the Catalyst can record an exemplar lap.
I would like to see the experiment done again with high-intermediate level drivers, and see what gains are made.
12/24/22 2:00 p.m.
I think for high-intermediate drivers it starts becoming important to look at driver input traces -- steering angle, brake pressure, throttle position, stuff like that. AFAIK the Garmin doesn't support that kind of logging yet.
12/24/22 2:40 p.m.
I think you can safely side-step the ethical problems by creating your own utopia, perhaps under the sea?
I was just more wondering if I could toss my catalyst in my students' car and go nod off for a session or two :)
12/24/22 3:12 p.m.
You can program a computer to play chess because at any one time there are a finite number of moves available.
While driving there are an Infinite number of possibilities - for which the human mind can draw on instinct, training or rules of the road to deal with instantly !
Given enough time and resources ( cpu / memory and good programming logic behind it) , an AI could continue to "Learn" things, but how long would it take to develop the repertoire of responces or intuition that some humans are born with or what We learn over a lifetime ?
There are crappy drivers and there is flawed AI, and they will be the Deat of us...
https://www.youtube.com/watch?v=yRdzIs4FJJg
Read Matthew Crawford’s "Why we drive"
PS: I used to be of the opinion that if self driving cars could save lives, then we could accept the future. But with self driving cars Killing People, I would rather put my life in Any Humans hands instead.
12/24/22 6:51 p.m.
Can you IMAGINE a Catalyst with a OBDII interface? Dang...
12/24/22 7:55 p.m.
Computers are good at precision, people are good at flexibility. Driving fast on a race track (time trial/qualifying, say) is all about precision, and they've gotten a long way in making automated cars do that. If they aren't faster than good drivers already, it won't be long.
Driving on the street is all about flexibility and IMHO we are over a decade away from a truly viable self-driving car. Frankly I don't think the current approach to it (so-called "deep learning"/neural networks) will ever get us there, we need a new technique.
Racing is probably somewhere in the middle, and a real-life race between the two might well come down to how aggressively the computer was programmed to drive. That would be pretty dangerous though, and I don't think it's likely to happen much.
12/24/22 8:35 p.m.
In reply to codrus (Forum Supporter) :
This reminds me of the time that a program noted the maximum possible lap time for an F1 car............and Ronnie Peterson went faster.
So there is this thing I do under real my heavy threshold breaking where by I preload the throttle ever so slightly to get the rear end to squat just before turn in.
I was talking with a former Indy car driver & IMSA front runner (I promised them In wouldn't ever use their name when discussing it these things as they didn't like to be quoted about driving technique) and he asked me where I picked up this technique. I told him it's a carry over form my motorcycle racing. Needless to say he was surprised an amateur had figured this out.
What's my point? Do you think it would ever occur to the person who's doing programming to do this? I personally doubt it, as my experience is that the people using the equipment and people programming it have very different mindsets.
12/24/22 10:50 p.m.
Indeed , Qualifying on an empty track , with thousands of test laps simulation no problem.
Add a field of 20 cars and you get the variables that make it a Humans game ...
12/24/22 10:53 p.m.
Ronnie Peterson died in 1978, you might have noticed that computers have changed a tiny bit since then. :)
As for the person doing the programming, if you're writing software for a racing team then absolutely. You get all the expert input from the drivers you can.
12/26/22 5:03 a.m.
I like that observation...but not the conclusion.
To be precise means the input variables are constant. In reality, they are not. Even in a single-car situation, there is no predictable perfect lap because the car's performance is changing constantly. Most notably, tire grip and feel.
As someone who pushes a variety of tires to the limits frequently, I can tell you it is a risky business even for a "flexible human". An AI isn't going to be able to predict the performance changes of a particular tire during a session without a LOT of test runs to base that curve on. Grip is not a single number. And that assumes fresh tires go on every session...which rarely happens. Instead, you have to know your vehicle really well and be able to sense changes in performance potential quickly and apply those to your future asks.
PS: GRM tire test stories never mention the offs and mulligans that occur in the name of science.
1/2/23 10:10 a.m.
In reply to Andy Hollis :
That subject alone, (thePS) sounds like a very interesting article!!!
1/2/23 10:17 a.m.
In reply to Andy Hollis :
Huh -I had not thought about the tire degradation piece, so add to that the changing track conditions due to temperature, humidity, that rubber wearing off /onto the track surface...
It just keeps getting more complex :)
1/2/23 1:52 p.m.
The computer doesn't need to predict the performance changes of the tire, it can measure them exactly the same way that you do. You do it by turning the wheel and sensing the resulting yaw and the feedback through the steering, the computer can do this with accelerometers and force sensors but it's got much more resolution. This isn't what I mean by "flexibility", it's just a closed loop control systems problem and we know how to build those.
When I talk about "flexibility", I'm talking about dealing with the potentially unending list of situations that come up out in the real world, rather than the controlled confines of a race track. The kid in the yard over there -- is he going to kick the ball out into the street and chase it? What about that man on the corner waving his arms, is he a crazy man or a police officer directing traffic? Construction zones with cones that make you drive over onto the wrong side of the road to get around a hole that they're digging. The list of those sorts of problems just goes on and on and on, it's effectively infinite. Humans can think, they can understand what's going on and figure out the best way to deal with it. Computers can't do that and we have no idea how to make them do it.
The race track environment tries to minimize these random factors, those that are left are basically down to the other drivers on the track or mechanical failures on the vehicle. Make it a closed single-car qualifying session on a permanent road course and you've got the ideal environment for a computer to go faster than any human can.
1/3/23 11:03 a.m.
Perhaps, but will the computer notice a cloud cover/uncover the sun and realize the potential change in tire grip later in the lap. How about seeing spectators raise umbrellas for a rain shower ahead. A human can.
Your proposition reminds me of the saying: In theory there is no difference between theory and practice. In practice there is.
BL
You'll need to Log in to post comments.