I've always wondered about why manual transmissions generally get better mileage than automatics. The answer is surprisingly simple: humans are better shifters.
According to the Environmental Protection Agency's fuel economy ratings, cars with manual transmissions typically beat their automatic peers by a mile or two per gallon. This is largely because manuals give you more control over an engine's exertions. Despite recent advances in slushbox design, humans are still better than automated systems at recognizing precisely when to shift gears. And smart shifting enables you to limit an engine's rotations per minute, which translates into less fuel consumption.
But here's my question: Why isn't there a computer that can shift better than a human? What's the unique cognitive talent behind gear shifting that a computer can't master? It seems to me like designing a "smarter" transmission would be a fairly easy task. If we can design a computer program that can beat a grandmaster at chess, or diagnose a heart attack better than a cardiologist, then why can't we build a transmission that knows when to go from second to third gear?
- Log in to post comments
I had a similar question some years ago (in 2000) when I was overseeing a seismometer at a field site. It was an old-school one which still had the drum paper output as well as a digitial version of the data. Part of my job was to change the paper every day and go over the record with a pencil marking the earthquakes. I would then go in and correct/adjust/add to the digital record of the quakes. There was software, honed for years by scientists at the USGS, to detect quakes, but somehow they were never able to quite capture what the human eye/brain can do without even thinking (I spent perhaps 5 minutes looking over a 24 hour record).
Actually, you're missing two fundamental points on why the autobox is less efficient.
Firstly, the automatic gearbox has a torque converter, which is no more than about 90% efficient until it's locked up. Most designs only lock in top gear, and above a certain speed.
Secondly, the human has an advantage: they can see the road ahead, and know what they're going to do about it, and can plan gear changes ahead of time. The computer controlled autobox can only react to the load and speed at the time.
I've always assumed this was true and thought it was very simple. The human doing the shifting knows how much he or she is planning to accelerate; the computer can only guess. Hence cruise control - any data on that?
Besides the slippage in the torque converter, automatic transmissions use hydraulic pressure to activate the clutches and bands to control the shifts and to power the control system. The pumps that pressurize and circulate the fluid for the hydraulics absorb quite a bit of power thus decreasing fuel mileage.
As mentioned above, the inability to predict future actions is the biggest difference. Without a video camera pointing at the road, best a computer can do is use past changes in steering and acceleration to predict what the driver is about to do. That said, probably the most efficient system is a completely computer driven car where the computer does know future changes and will probably accelerate and break more smoothly than most people.
Also, I always thought another advatage of manual transmission was weight. I'm not sure how true this is, but the extra hardware for automatic adds enough weight to have a small but measurable MPG change.
This is all very interesting to me. Thanks for your astute comments. Obviously, an automatic transmission can't forsee future road conditions. But can't someone build an algorithm that can predict the need for future gear changes based on recent driving dynamics? I'm thinking that there has got to be some way to interpret a driver's actions so that a computer can tell if the driver is in a city on merging onto a highway.
Hmph. Quibbling about 2mpg transmission differences? How's about getting rid of the steering wheel and letting the computer(s) control navigation? That could produce some significant reductions in fuel consumption. (Insert ominous "Sorry, Dave, I can't let you go there..." here.)
I doubt seriously that a typical driver is better at choosing a gear than an automatic. That might (or might not) be true of highly skilled drivers (and who among is is not highly skilled?), but not of the average joe or judy. I think virtually all the increase comes from higher mechanical efficiency in the direct linkage of a clutch versus the hydraulic linkage of an automatic unless it's locked up. And, along those lines, there are some vehicles that get higher fuel economy on the highway with an automatic than with a manual transmission, in most cases because of a higher (lower numerically) final drive ratio. Around town, where most of us drive most of the time, the automatic spends more time being slushy, while the manual spends more time with its drive train locked mechanically.
I remember an article discussing Deep Blue (or whatever that computer was called) and comparing how it functioned to how the player's brain functions. The article pointed out the Blue had a massive assembly to cool it and was/is one of the higher-end supercomputers. It worked by simply postulating every possible move, then selecting the best one.
Humans, on the other hand, work differently. Instead of working out every possible solution, we use past experience filtered through a stochastic algorithm to postulate the optimal result. While not a precise as the brute-force method, it has advantages in it's lower proccessing overhead and faster response time.
In the transmission, humans can select the proper gear faster than a computer. This is because we use not only a different algorithm, but a different type of math altogether from a computer. Thus humans, and indeed all thinking organisms, are better at manuvering through space than a computer of comparable power. Only when we have a super computer in a car will an automatic transmission be as capable as a human.
Perhaps one also could investigate stats on who chooses manual vs automatic, and why? 2mpg is about 10% for a city consumption rate of 20mpg, typical. Equivalent to the non-cruising slush loss of the automatic gearbox. Which suggests that the human's forward looking behavior could be as often "wrong" as not. I suspect that the automatic algorithms are probably as good as a humans' when considering getting from A to B style driving with no emergencies or other drivers to avoid.
My feeling is that stick is however far safer when driving at or near the limits of adhesion, whether by choice or by emergency, since automatic gear-shifting under power at the wrong time can shock the system and break adhesion. No computer is going to have the senses for that kind of situation.
So says one with many 100K's miles under his belt and a mis-spent youth exercising the above choice on the Santa Monica mountain roads. I still miss that first-version stick-shift VW Scirroco of the '70's. Sigh.
That could also be bad. If the computer guesses wrong, you might fly through the next intersection or sit dead on the freeway. (The way I see people merge, I can understand why a computer might get confused.) It's probably safest to keep it naive.
Isn't there also the matter of the number of gears? Most manual transmissions have a 5th gear, so you should be able to minimize the amount of work done by the engine, especially at high speeds.
Some manual transmissions have six gears, but then so do some automatic transmissions. The automatic and the manual might be heading towards a merger, somewhere between CVT (constantly-variable transmissions) and automatic clutches.
I'm reminded of a story told to me by a friend, who works with computational mathematics. One day, his boss gave him an assignment to write a program to solve a particular problem. As he looked over the requirements, he realized that it was impossible, and informed her that the problem couldn't be solved with a finite state machine. She told him to just build an infinite state machine.