Robustness Through Inefficient Design

Friday, August 28, 2009

Early this month I heard about the ECCERobot, or Embodied Cognition in a Compliantly Engineered Robot. While standard humanoid robots mimic humans in appearance, ECCE copies the internal structures and mechanisms. The anthropomimetic design incorporates such elements as bones, joints, muscles, and tendons, allowing for the potential for human-like activity. ECCE’s drive system of elastic cables and drill motors may be complicated and inefficient, but it provides insight into our own design.

It is this same complexity and inefficiency that differentiates us from the machines that we design. In general, we try our best to design components to be as simple as possible. The benefits of The benefits of simplicity are threefold: firstly, simple components are cheaper to manufacture and maintain; secondly, simple components are often more efficient (but not always); finally, simple components are easier to understand and operate. Although the tools that we design have progressively grown more complex to fit our needs, many existing designs have been drastically simplified. For example, watches used to contain carefully constructed clockworks of gears, weights, pivots, and springs all working together in synchrony. Gradually, battery-powered motorized watches replaced the old-fashioned mechanical designs, eliminating the tight constraints that made these older devices costly to produce. Nowadays, digital watches run by integrated circuits have become popular. The ingeniously designed clockwork of yesterday has not disappeared altogether, but it is not nearly as common. While integrated circuits are arguably more intricate than mechanical and motorized methods of keeping time, they are constructed of repeating elements and trivial to mass-produce. And compared to these older designs, digital watches have far lower power requirements.

The same could not be said of us, or indeed of most other constructs found in nature. The simple motion of lifting an arm involves the perfect cooperation of numerous bones, muscles, and connective tissue. Neural signaling and processing is centered in the brain, an organ whose sheer complexity we have only begun to grasp, but whose number-crunching power is vastly inferior to that found in our microprocessors. Language and communication between individuals is inefficient and inaccurate, relying on assumptions and sometimes translations to convey even the simplest of ideas. It is amazing how simply smiling would utilize most of the facial muscles, whereas the same message could probably be contained within a datagram of a few bytes.

Our inefficiency and complexity is not a weakness, however. While it is true that our machines are highly optimized and therefore perform specific tasks extremely well, we have a robustness that our designs lack. All it takes is a single loose screw or solder joint to render a watch inoperable. A single broken transistor may render an entire computer chip unreliable. And if a datagram is even slightly altered due to noise, the entire message may be lost. On the other hand, our complexity makes us inherently redundant. Though we may become ill or injured, our illnesses and injuries do not affect us the same way that damage affects our machines. Our performance may degrade as a result of injury, but the degradation usually scales to the degree of injury. We try our best to design redundancy into our creations, but they are still very susceptible to Byzantine failure.

The ECCERobot may be inefficient and complicated, but it also inherits our robustness. By copying our own mechanisms, we try to create a robustness that is difficult to design from scratch. But where did our own design originate? Our structure does not appear optimized; indeed, it does not imply a purpose at all. We were not designed, but refined through eons of selection.


  1. Jenny says:

    This post is incomplete :p

  2. Jiang Yio says:

    Thanks for the heads-up; it’s been fixed.

  3. Primefalcon says:

    Very interesting article, The digital watch however also performs far more functions than any clockwork mechanism could….

    But yes from a design perspective I agree robustness and efficiency will more than likely win out in even the robotic designs.