Automotive Neural Networks - Pt. 3 - An Equal Relationship With OEMs
Part 3 - Our Role In The Future Of Automotive
See Part 1 Here
See Part 2 Here
Up to this point we have briefly examined the construction & purpose of artificial neural networks (ANN) in the automotive field, as well as some of their early and current uses. Now we can examine the exciting part: their future in automotive and our role in it.
OEM's are continuously experimenting with neural networks as well as other types of Artificial Intelligence (AI) in order to more efficiently design, model, control, and diagnose the ever increasing number of systems contained within their products. Research & development divisions spend countless hours and dollars trying to do this, but there is a problem: modeling and training in laboratory conditions is not nearly close enough to real world conditions to make these systems as precise as they need to be. That is one of the reasons that we have not seen an overwhelming switch over to these networks; least not that I am aware of.
Real world data is needed to make these neural networks more efficient and actually useful for the end user. Seems simple enough, but for any manufacturer it incurs quite a bit of expense to do this. The countless software/hard part updates every vehicle of every manufacturer has had is testament to that. There is also the time factor.
This is where we as diagnosticians & technicians come in. We are collectively sitting on the largest compilation of real world data in the world. Every time you complete a repair order, every time you take a scope capture, every time you document a vehicle with a case study, you are contributing to this library of data. Big data as it has commonly been called, is soon to become our biggest bargaining chip. Our collective library of data and knowledge will be exactly what we can leverage to begin to change the industry to a new model that better benefits all involved. The best part is, all that is our data to negotiate with and the manufacturers severely need it, even if they don't know it yet.
Sounds like a pipe dream doesn't it? The big corporations don't care what us little guys think right? Wrong. I firmly believe that the biggest disconnect between the OEM's and the repair industry is simply a lack of communication, or perhaps more accurate is the fact that we are both speaking to each other but it seems to be in different languages. In the future, we will both need to co-operate fully in order to maintain vehicles at the level the consumers will expect and require. This will consist of a mutually beneficial (and properly regulated) exchange of information, tooling, and techniques. There will need to be industry panels, discussion, negotiation and the like by various representatives of our field but that will all come in time.
Picture this: a manufacturer is designing an updated vehicle begins with modeling of the various control systems with laboratory training data. Then they feed this initial model real world data: our known good and known bad waveforms from various sensors/modules/components. Case study data with failures and confirmed repairs and so on. This allows the system to be able to recognize and more effectively self-diagnose, or at the very least more accurately report failure details (freeze frame). Using neural networks may even allow the interconnected systems to make changes within parameters to maintain emissions, power, operational, and comfort expectations until such a time as the vehicle can be properly repaired. There was a white paper released describing self healing automotive systems (self healing in the sense that it detects a malfunction and fixes it through software tweaks or through actual manipulation of electrical components and their operating parameters.)
A library of known good and known bad is a great start, however, we all know there is so much more to automotive failures than just failed components. Imagine having all the experience we collectively have in the form of a database. The knowledge that says euro cars tend to be fish bowls (water intrusion), poor wire harness routing locations that causing chafing and intermittent shorts, common water leakage points, actual end user treatment and operation of the vehicle. All of these can be modeled, to some degree or another, in a format that will allow these neural networks/AI systems to start to see vehicles in the holistic manner that we do. This allows the manufacturer to take all this into account during design and building, but also during diagnostics.
Snap-On has recently released their "Zeus" scan tool with intelligent diagnostics, something that I'm not very impressed with at this stage. I have spent many years mentally playing with something much more complete and useful. I envision something along the lines of an augmented reality set-up, where highly trained technicians, diagnosticians, and neural networks all work together to repair vehicles. A Iron Man type HUD (heads up display) {though hopefully not a full helmet, though I wish I had one some days} that is factoring in vehicle condition, fault codes, wiring, component interactions, and so many more parameters. Caterpillar and Bosch are two companies that I know have been experimenting with this and I think the training and service applications are awesome. A diagnostician/technician inputting additional data or modifying the data the AI is putting out. Teamwork with technology at an OEM level. I envision that we will have our own neural network systems that we use as another tool alongside our scopes and wrenches.
I see us being able to connect to a vehicle and actually start to break down the components of these systems, to be able to actually throw inputs and variables in so that we can see how the network is interpreting them. Feed in "black swan" data (data or operating parameters that the system has never seen before) and document the results. We can then take our known input, see our outputs, and reason (both deductively and inductively) what is occurring within the "hidden layers" of the system. Skilled diagnosticians/technicians will be invaluable at that point. We will, in a very real sense, be tailoring each and every vehicle to it's end user.
As this modeling/data cataloging continues it won't put us out of work, even when vehicles can effectively self-diagnose. In fact it will put us more in demand. There will be a major drop off of the people in the industry who don't wish to invest in training, equipment, and proper compensation for employees. Those that remain will reap the benefits. We will be considered research scientists and we will actually spend a lot of our time actively improving the operation of vehicles right then and there in the bay, instead of waiting for someone else to do it.
If anyone wants to see some of the possibilities that lay ahead, read some of the white papers from NVIDIA. Their technology seems to be at the forefront of enabling this future, especially with their Tegra X1 chip and the CX/PX platforms. I believe that there is probably at least one member on the forums here who might even be able to get some representatives from NVIDIA or OEMs in on the discussion.
Thank you all for taking the time to wade through all this. I know I am rather optimistic about the future but I'd rather see the good that could come than sit here fretting about the possible bad. Leave some feedback, thoughts, discussions, Y2K conspiracy theories, whatever you may have.
I am going by my gut on this one. I skim read a lot of this stuff about AI etc., but my hardware is not up to the task of full comprehension. I agree that those that are not invested are out. I disagree that we will be invaluable skilled diagnosticians. I believe the cheapest , least skilled labor capable of performing the tasks needed will get that job. I believe the "device" will be a
Bill, Exactly kind of thought I was looking for. I see your points and can't dispute them with any hard facts. Sadly, I believe given the state of things you are probably closer to correct than I am. Your scenario sounds like the movie THX 1138. I'm going with a more utopian view, plus I believe if the full benefit of interacting with out of warranty vehicle data is demonstrated and leveraged
Hi Chris. I'm a realist and have been a mechanic turned technician for the past 50 years, so I've seen plenty of change and an extremely slow development and incorporation of technologies that were first drawn up by Leonardo Da Vinci. So, personally I think that you're viewing this through rose-tinted glasses, with a healthy dose of wishful thinking. The whole concept of virtual reality where
Martin, Rose colored glasses and wishful thinking is a polite way of putting it, I won't argue that at all. The funny thing is in almost all aspects of life I'm a pragmatist, to a fault even. There are certain things, such as our industry actively taking control of emerging technology and using it to our advantage, where I become rather (read as overly) idealistic. I dont disagree with any of
Hi Chris. You know, we've been promised the very changes that are being discussed here for at least the past twenty years, so at best the applications have been slower coming than we were at first led to believe. I can certainly see some value in the changing technologies, but until costs become viable some these aids will remain on the outer boundaries of the field. I've always been open to
That's interesting information Martin, I was not aware of that GM program, definitely was a step in the right direction. Growing up I was pushed heavily into engineering but I diverged from that path. I decided to focus my education on repair and diagnosis instead of design. I have a great respect for engineers still. Still have my collection of Audels handbooks, and my favorites How Things
Hey Chris, Your optimism is great and I'm sure that will serve you well into the future. I try to stay positive too but sometimes I think I've seen too many dystopian type movies over the years and I dwell on the negatives. Bottom line is that none of us can predict exactly what things will be like or how things will work out in the future. The one thing I am pretty sure of though is that we
I think you've overlooked the biggest part of the future of the automobile - autonomous operation. IMO, autonomous vehicles will eventually kill private vehicle ownership, especially in urban areas. Better to pay a small monthly fee for a car service rather than own a vehicle that'll be sitting in a garage or parking lot 90% of the time. No private ownership means in-house fleet care by the
Chris, As I stated above, dissenting opinions are solid gold to me and I appreciate everyone who is taking the time to respond. Once again, I agree with the vast majority of what your saying. I still see an independent market being a viable option as I still believe our skills will be required, even if its just the bleak landscape where we are working on fleets that have been turned over once
Interesting stuff. Just read quickly through the three parts. I have yet to watch any of the videos etc. but I will. I am with you in thinking that our skills as a group need to be put to work to advance our trade into the coming technologies. We have to leverage our position as diagnosticians and get our hands/minds into the diagnostic processes of the future. ANN in the directed format will
I appreciate the insight and the addition to the conversation. If nothing else, it's a discussion that is perhaps better begun now rather than later. If you run into any questions in anything let me know, I'll do my best to clarify, or direct you to information that does it better.
I'd like to add my two cents. When thinking about proliferation of new technology, I like to consider two things: economic incentive and regulatory hurdles. - When it comes to ANNs in safety critical systems, it is not obvious how to certify them. For example, if an airbag system has a mind of its own and "learns" something that may stop it from triggering under conditions for which a new car
Dmitriy, I like the pragmatism of your thoughts here. Not overly invested either way. It is a good addition. I think it will be years (if at all), before we seen ANN in safety systems such as airbags. Though an arguement could be made that ADAS is exactly that in a way. I don't think self diagnostics needs to be a selling point, in the same way that OBD2 or multiplexing isn't a sales point…
Hi Chris: I'm a bit late to the party. (Life gets in the way at times.) Brain-pain concepts like this requires a certain amount of free time to digest. There have been some interesting thoughts in some of the responses. I think that it is important to know where you've been to know if you're going forward. Like Marty (and Scott though he hasn't chimed in), I've seen a bit evolution of these
Anthony, That was excellent input and I have to research more into all of it because you brought up multiple points that I wasn't aware of, including Christian Borgy. I was hoping to stir up these kinds of debates and draw out some of the vast knowledge lurking around here. I've gotten alot of info and view points that give me more room to explore. I think I've said it too many times
Hi Chris: Interesting TEDx Talk though I was a bit disappointed with the bridge. In the talk, you're led to believe that the bridge is being 3D printed across the canal. That would've been impressive (a self built-in-place cantilever bridge). Alas, that's not the …e. curbed.com…I'd love to see the testing on the race car frame. I'm thinking the testing is why
HI Guido. We all know who bore the cost of beta testing SI! It was the technicians. From the first iteration of CD and server-housed discs that was not a Windows driven program, it was a nightmare, through the SI2000. The learning curve wasn't nearly as bad as the development and implementation of the infrastructure! While all of that was a very expensive disaster that was dumped in favour of
Hi Chris, Here is a quick video on the pace of AI development. In the video there is a description of "Generative Adversarial Networks" that are pushing the pace of AI development forward. Pretty interesting stuff. youtube.com/watch?v=mQO2Pc…
youtube.com/watch?v=OEu4Iq… All kidding aside, there are some serious implications involved. The ophthalmologists decision-making should NOT be surprising. It's been long proven that judges' decisions vary based upon the time of day. The noted economic disruptions and how they are handled are definite issues or else the long lost "family" of Ned Ludd will be back and taking to the streets
The ophthalmologist finding has been found in many other "experts" in their fields as well, including judges as you mentioned. What I find fascinating is who decides that the output of any ANN or AI is in fact "good"? That is an ongoing debate. When an expert disagrees with themselves within hours of their diagnosis, what is to say that they can accurately gauge the output of a computer as
Hi Chris: I fixed the omission of the word "not". Since you knew about the judges, you must've known that the word was missing. Many, if not most situations are multi-variable. That's why dumbing things down is so dangerous. Dumbing things down allows FUD and those who spout it to fester. AI Self-Destruct, huh? Well, Ned's Nitwits have been known to wear down superior intellects…
Bob, GAN networks are quite fascinating, but even that has it's own limits. In reality, as we move forward, you will continue to see GAN paired with various types of neural nets and deep learning techniques. There is no one size fits all solution, even using different transfer functions in the network drastically changes the results, using a sigmoid function vs a step function or linear
Hi Chris: I like how he told NHTSA to go pound sand. Actually, it was in what he didn't say that I like. "The comma one is cancelled. comma.ai will be exploring other products and markets. Hello from Shenzhen, China." There's not much they can do about it. (Note: This was 2 years ago.) Smart kid. No one gets to see his stuff. I know. No one at NHTSA would ever share IP with
That's really cool. It seems like many of the big discoveries and inventions in the world end up coming from some unknown guy toiling away in their garage. One of the things Peter Diamandis talks about is the confluence of all the emerging technologies and how they build off each other. The other big thing is how many people around the globe are coming online and having access to the sum total
George Hotz is definitely interesting. It is always quite exciting to see where the next breakthrough is coming from. Sadly, even with the compilation of humanities knowledge readily accessible, there are so many who don't take advantage of it. I have a glass of bourbon in salute every time I hear that another library has closed. We often don't take the knowledge available to the next level and