|Developer Diary · You Heard It Here First · 28 December 2003|
|Robotics Profits Elusive|
Despite growing sales profits in robotics are elusive. The gradual pace of adoption seems to be most responsible for the slow financial returns from the sector. Of course there are some companies that are growing rapidly. In particular it seems to be the companies making the simplest systems such as inventory and packaging systems. A typical example that is more visible to the average person is Optimal Robotics, a Canadian company that makes automated checkout systems. You may have seen these in supermarkets or at Home Depot. They allow a single clerk to man as many as four checkout lanes while customers scan their own purchases. It hardly seems fair to describe self-checkout as a robotics application, but these are the kinds of systems growing the fastest.
The larger question is if and when robotics will break out and really become pervasive. Though checkout systems and restaurant ordering kiosks may fascinate the public when will automated motion really translate directly into industry? After all we have had ATMs for twenty years and still robotic growth is average. Marshall Brain has made some predictions about adoption in a series of articles such as the recent one entitled Robotic Nation. His luddite attitudes about robots somehow robbing people of their valuable "jobs" is a typical American attitude responsible for holding back growth in America. Nevertheless he does correctly point out that machine vision is important to the advancement of robotics.
I would go even further than this in stating that the primary reason why robotics growth is steady instead of explosive is because of the sensor bottleneck. The lack of development and integration of sensors is impeding the practicality of most advanced robots. The canary test case is Cognex located right here in my own backyard of Natick, Massachusetts. Cognex revenues have been plunging and they lost money for the first time last year despite being the industry leader in machine vision. This year they are doing much better as capital investments increase. The real story here is to look at Cognex's customers. Most of their customers are wafer manufacturers for the semiconductor industry. What this tells you is that machine vision is so incredibly complex and expensive that it can only be used by the absolute richest manufacturing applications.
Analyzing this a little more: a high-speed camera costs $1,000 or less and ditto for accompanying computer. In fact if you skimped you could probably pay less than $500 for all the hardware in a vision system. So why do these systems cost $50,000 a pop? Answer: integration and programming costs. In this area of robotics above all others you see the huge software problem.
Engineers in the business call it "machine vision", but in reality what you have here is a problem in AI. Capturing images is easy. It's interpreting and using those images that is hard or impossible and this is what is holding back robotics. Also, it is not just a question of vision but of all sensor types. This lack of vision not only limits the capabilities of robotic systems but perhaps more importantly it enormously retards and complicates what could otherwise be everyday tasks.
Using a robot is bit like hiring a blind machinist. A blind machinist (were there such a person) is more at risk for being hurt or causing damage, works much slower, cannot do many kinds of tasks and takes an enormous amount of training beyond what a sighted machinist would require. This is why there are no blind machinists. Yet in robotics this is exactly the situation; near universal blindness due to lack of intelligent sensor integration.
If there will be a breakout in robotics it will likely come when reliable machine vision becomes a reality.
|return to John Chamberlain's home · diary index|
|Developer Diary · email@example.com · bio · Revised 28 December 2003 · Pure Content|