Monday, October 30, 2006

More PRT Disinformation

Avidor recently reprinted a comment from David Greene, a software engineer from Minnesota who doesn't think PRT can work because the computer algorithms are too complex. As a fellow software engineer, I feel compelled to respond to his post, point by point:
Greene: "The idea that computers can automatically control a distributed network of machines that by their nature have a human interaction component and that this system will work at peak efficiency with vehicles traveling inches behind one another is absurd."

What is absurd is Greene's statement that PRT vehicles would be traveling "inches behind one another". The truth is, vehicle separations are measured in dozens of feet even for very short headways, with conservative designs exceeding one hundred feet. See "PRT is a joke" is a joke" for an expanded discussion of vehicle separations for different systems.
Greene: "We have plenty of examples of much smaller and less complex systems have taken years to get right, if at all. A few that
come to mind:

- Denver airport's baggage handling system

- MSP airport's trams (which only go from one single, fixed point to another!)

- Metro Transit's GoTo card system"

None of these systems has anything to do with PRT or PRT companies. It's like saying light rail is doomed to failure because of Amtrak's financial troubles. Amtrak and light rail are at least as comparable as PRT and the Denver baggage system.
Greene: "As a computer professional, I can state with utter certainty that computers are not the magic bullet that will save us. At one level they are extremely simple machines. But at another level, they are incredibly complex. Debugging distributed software is no easy task. Programs will always have bugs. The question is, can the developer get rid of enough of the critical ones to make the system mostly functional? Is mostly functional good enough?"

First off: I never trust someone who makes predictions with "utter certainty".

As for his statement about complexity, he neglects to mention the fact that very complex distributed systems are all around us - so much so that we take them for granted. Some examples of massively distributed systems that we use every day without even thinking about it:
  • The Internet - over half a billion computers on a heterogeneous worldwide network - and any one can connect to any other within milliseconds
  • ATMs - hundreds of thousands of fully automated kiosks, providing financial services to millions of users. It would seem that they've worked out most of the critical bugs, despite having to integrate real time financial data from hundreds of different financial institutions.
  • Inventory systems - retail giants are able to track every piece of inventory in every single store, with such accuracy that an online shopper can check store inventory, purchase an item online, and pick it up in-store within 15 minutes.

Compared to these problems, routing a few thousand vehicles on a local guideway is relatively simple. Note: I didn't say absolutely simple - certainly there are complexities in controlling a PRT system. But if we can tackle problems as complex and distributed as the Internet, ATMs, and global inventory systems, is it really such a stretch to assume we can handle PRT?

But in actuality, no assumptions are necessary. Morgantown PRT, though not technically a true PRT system, has featured automated control of vehicles on a switched track for over 30 years, without incident; Cabintaxi was a thoroughly tested German system which was able to route two dozen vehicles over 1.9km of guideway with 6 stations. Both of these systems were built in the 1970s, when computing power was measured in kilohertz. More recently, ULTra has logged thousands of miles of testing at its Cardiff, UK test track. All of these systems have carried passengers.

In other words, it is absolutely absurd to "state with utter certainty" that PRT computer control will not work - since it's already been demonstrated to work by multiple independent groups.
Greene: "In technical terms, the problem of optimally scheduling a distributed network of traveling objects is likely NP-complete (I haven't done the formal analysis). That means that the computer will have to make guesses and I can guarantee that it will guess wrong in some cases. That doesn't mean the system will break down, but it won't always achieve the peak efficiency that its promoters proclaim."

This is mostly flat-out wrong, despite Greene's "guarantee". PRT routing and control algorithms are among the most studied and simulated aspects of PRT research. In addition to the hundreds of thousands of miles of real system testing done on Cabintaxi and others, in-depth simulation studies have also subjected these routing algorithms to billions of miles of testing, validating the algorithms under the most extreme loads.

Modern algorithms feature asynchronous control, which is decentralized, scalable, and flexible. Vehicles can be largely autonomous, especially when it comes to maintaining safe distances, so regional controllers are not burdened with every safety detail. Central control can focus on global optimizations to optimize routes and reduce congestion.

Is every single passenger guaranteed an optimal ride? Of course not - no transportation system can make such a guarantee in heavy loads. But extensive analysis, simulation, and testing have shown that these systems will operate at or near optimality for all passengers.
Greene: "A system like PRT will need many layers of redundancy, and that adds cost."

Nonsense. Computers are cheap, as are most of the mission critical components that would require redundancy. The vast majority of the cost of PRT is in guideway and construction materials, and in the manufacturing of the vehicles. Adding a few redundant components to such a large infrastructure only adds a trivial amount of cost.
Greene:"I just don't buy the claim that private companies will pay for it and make a profit. If that were the case, why does PRT need any subsidy at all?"

Light rail requires subsidy for both construction and operations. That doesn't stop us from building more light rail. PRT, on the other hand, offers the promise of providing much better service (try taking the train at 3:30am on a Sunday morning) for less operating cost, and may even be able to recoup the construction costs. At worst, if all the projections are terribly wrong, we'd wind up with just another subsidized system like LRT - only more convenient and available 24x7.

Private companies haven't embraced PRT for the simple reason that it's too much of a corporate risk to sink hundreds of millions of dollars into. Big companies, like governments, are wary of any radical challenge to the status quo.
Greene: "I work in the supercomputing industry developing the software that users need to develop their applications. I also do some high-level hardware architecture work. A system like PRT will need something equivalent to a low-end supercomputer to work. I know how difficult it is to get these systems to function."

This is absolutely, ridiculously false. A low end supercomputer? Come on! The most complex PRT computations are routing and trajectories, neither of which requires anything near a supercomputer. Computers have been calculating trajectories since World War II! Heck, an iPod could probably handle the load of a small PRT network!

As proof of this, I again reference Cabintaxi - which was fully controlled by computers five years before the original IBM PC came out!
Greene: "The claims made by PRT proponents are pie-in-the-sky nonsense. They are made by people who have no real understanding of the underlying technology needs and complexities."

The only nonsense here is Greene's wild assertions. He is either hopelessly misinformed, or is just another mindless follower of Avidor's propaganda campaign. As a fellow software engineer, I'll give him the benefit of the doubt and assume he just hasn't done his homework.


Post a Comment

Links to this post:

Create a Link

<< Home