Working Assumptions

July 2014
By Bryan Bergeron

Designing robots — especially long-lived ones — involves making numerous assumptions. For example, there's the availability of a standard power source such as AA batteries or a 11.1V Lithium battery pack. Then, there's the operating environment. If we're talking carpet roamer, then it might be shag carpet up to, say, two inches thick.

Assumptions get more involved when there's an intermediary involved, such as a tool that the robot uses to interact with the environment. For example, a wrench used by NASA's Robonaut 2. Things get even more interesting when the environment involves a human — even the same human. The obvious example is a surgical robot.

Let's consider that our robot performs the most elementary prep for, say, abdominal surgery. The first step — prior to applying a drape and antibacterial coating — is to shave the area that will be incised and sutured. So, let's say you're designing this shaving robot. What are your assumptions? For one, you'll have to consider exactly what type of razor the robot will be wielding. Should it be one of the new four-bladed varieties or should you "design ahead" for, say, a five- or six-bladed model? It matters because of the surface area covered, and the drag produced by the blade assembly will vary from one design to the next.

Let's say you opted to stay with the currently popular three-bladed shaving head. Now, let's look ahead five years. What if it turns out that the multi-blade shaving fad has given way to less expensive "retro" double-edged shaving? Double-edged (DE) blades have less drag than their multi-blade counterparts and don't clog as easily. They also require more skill to use without drawing blood. Following this scenario, your robot will likely need to be reprogrammed and perhaps fitted with additional sensors to make up for the generally easier to use multi-blade shaving head.

From a personal perspective, I've recently moved back to DE shaving. It required me to relearn skills that I haven't used in years, and I've lost a bit of blood in the process. Plus, I'm human. Imagine what a surgical robotic assistant — even one that can learn from experience — would have to go through to learn how to use a new razor or other surgical prep tool. I wouldn't want to be oneof the first dozen patients in the operating room of the future.

One of the characteristics of expert systems is their inability to operate outside of fairly constrained domains. The better the expert system, the less likely it is to "blow up" as the fringes of its knowledge are approached. Similarly, for a robotic system, the better the robot — including onboard AI — the more likely it can adapt to a new situation and to a new tool set without failing.

So, what's the solution? One is to simply stockpile whatever you think your robot will need in the future — from spare parts to spare tools, whatever those tools happen to be. The other is to design the tool handling interface so that it is as general as possible. Don't design for a particular make or model of a tool, regardless of how popular the tool happens to be. In this regard, you need to understand the tasks that you're asking your robot to perform. You may not be able to predict the future, but you can at least review the historical record and accommodate the recent past. Who knows, the barber robot of the future may be giving shaves with a singe straight edge razor. SV