Team4 ROBOT Dexterity

The researchers first had humans wearing motion capture equipment perform a variety of simulated everyday tasks, like stacking cups, opening containers and pouring out the contents, and picking up items with other things balanced on top. All this data — where the hands go, how they interact and so on — was chewed up and ruminated on by a machine learning system, which found that people tended to do one of four things with their hands:

Self-handover: This is where you pick up an object and put it in the other hand so it’s easier to put it where it’s going, or to free up the first hand to do something else. One hand fixed: An object is held steady by one hand providing a strong, rigid grip, while the other performs an operation on it like removing a lid or stirring the contents. Fixed offset: Both hands work together to pick something up and rotate or move it.

One hand seeking: Not actually a two-handed action, but the principle of deliberately keeping one hand out of action while the other finds the object required or performs its own task. The robot put this knowledge to work not in doing the actions itself — again, these are extremely complex motions that current AIs are incapable of executing — but in its interpretations of movements made by a human controller.

If robots are really to help us out around the house or care for our injured and elderly, they’re going to want two hand at least. But using two hands is harder than we make it look — So this robotic control system must learn from humans before attempting to do the same.

The word "dexterity" is thrown around a lot in the robotics world. It's used by robot manufacturers who describe their robots as dexterous. It's used by robotics researchers who describe the dexterity of their developments. It's used by gripper manufacturers who describe their grippers as dexterous.

You might ask: How much dexterity does my robot need to have?

The answer: It's impossible to say. Nobody can seem to decide what makes a robot dexterous.

Robotic dexterity is a complex topic. It's hard to tell how much dexterity you'll need for a particular robotic application — or even if "how much" is the right question to be asking.

Robotic dexterity = a moving goalpost
To understand why this is a tricky topic, it's useful to look at an example.

Last year, a research team from UC Berkeley claimed that they had created the "most dexterous robot ever created".

When I heard the news, I was skeptical. Not because I doubted that they had made a technological breakthrough. I was skeptical of the word "dexterous". I knew from experience that this word is a moving goalpost in the world of robotics.

You see, I investigated robot dexterity as part of my PhD, which I completed back in 2014. During my research, I discovered that there is no standard definition for "dexterity" in the robotics community. As a result, researchers often claim that their robot is "dexterous" without defining what this means.

This makes it very difficult for robot users.

If we don't know how dexterous a robot is, how can we tell if it's the right one for our task?

I was right to be skeptical. The UC Berkley team had done what many researchers have done in the past. They had invented an entirely new metric to measure dexterity. Their metric ignored the physical properties of the robot and instead focused on machine learning performance. There's nothing inherently wrong with their new metric — which is really a measure of bin picking speed — but it certainly can't be used to prove that the researcher's robot is "the most dexterous robot ever created." In order to say that, you'd need to measure all dexterous robots using the same metric.

This is not an isolated case. I've seen the same thing happen again and again. People use their own definition to try to prove a robotic system is dexterous.

There is no standard metric for dexterity — not until ISO nails it down at least. Unfortunately, that means you have to find your own way to assess a robot's suitability for your task.