A Software Engineer’s Solution to Robotics
Why General Purpose Computing Dominates
Mechanical and early electrical computers started out as purpose-built machines for each job. Each equation or logical operation had to be physically built into the hardware. As computers have gotten more powerful, they’ve also gotten harder to design, and the production facilities have gotten more expensive.
The marginal cost of making one more chip, even on the cutting edge, isn’t that high, but designing a chip and setting up the manufacturing line is so expensive that it’s hard to justify a new one unless you’re going to make tens of millions of them. Those extreme economies of scale for computing hardware have driven a paradigm where the hardware is general-purpose, and almost everything is a software application.
The Allure of Humanoid Robots
The allure of a humanoid robot is that, in principle, it can do anything a human can do, and our world is built for humans, so it should be able to do anything. They may not be the best at anything, but robotics “evolving” from specialized machines to a general-purpose chassis where every application is software would mirror the development of modern computing. That makes humanoid robots an extremely compelling vision and the logical endpoint of robotics to people who view design through the lens of modern computers.
Robotics isn’t like Computing
The Independence of Components
The first mistake in viewing general-purpose robots as the future is a fundamental misunderstanding of where the economies of scale live in robotics. For a computer, the hardware is an integrated and largely indivisible unit, so the only way to make 10 million of a component is to make 10 million of that computer unit. In a robot, the computing hardware, sensors, and motors are expensive and need huge economies of scale, but the chassis to put them in is comparatively cheap and easy to design and build. A dishwasher, a washing machine, and a dryer, for example, do very different jobs, but the same computer chips, buttons, motors, and pumps can be used to build all three. Specialized robots don’t require specialized components, and the important economies of scale are in the components.
Depreciation and Idle Time
The second big mistake is not understanding the differences in depreciation and the cost of idle time between robots and computers. Computers depreciate quickly because their speed and energy efficiency improve quickly, so replacing a computer with a newer model becomes cost-effective relatively quickly. That depreciation makes idle time incredibly expensive and makes things like flexible cloud computing that minimize idle time so valuable. Robots don’t depreciate the same way because they do physical work, where speed and efficiency don’t improve at nearly the same rate, if at all. That makes specialized robots that are better at their jobs but don’t get used all the time a much better value proposition than doing the same with computers. And unlike a single general-purpose system, that set of specialized robots can all do their jobs at the same time.
Why We Use Tools
Humans use tools because there are a lot of things we just aren’t very good at. A robot that mimics the human form will inherit the flexibility of humans to do different tasks, but also the same limitations. Humanoid robots will need tools to do most jobs the same way humans do, and at some point, you have to ask if it’s better to just turn the tools into robots than to make a separate robot to use them.
Leave a Reply