Elon Musk (among others) has been promoting the idea that robots with human-like abilities will be available within a couple of years. There are reasons to think that it may actually take a couple of decades, if it ever happens. The main problem is that robots (including Tesla’s) do not have human-like dexterity, and won’t for a long time.
My argument for robot limitations is based on reading the observations of those who are working in the field and who know what robots can and can’t do.
I really got interested in learning more about the state of robotics when I read a blog post by Rodney Brooks. A former MIT professor, Brooks is the founder of iRobot (the company that makes the Roomba robotic vacuum cleaner) and several other robotics-related companies.
In the blog post linked above (which is both very long and somewhat techy), Brooks argues that progress in the areas of touch and dexterity has been much slower than progress in text, vision, and voice understanding. That’s in part because robots don’t have the subtle touch sensors we have—there are thousands of sensors in our finger tips, plus hundreds more in the rest of our hands and in our wrists and elbow joints. Most of those are missing in robots.
We also don’t have systems for recording and playing back our sense of touch. That’s something we need if we want to train robots in a comparable way to the training of AI systems. We do have recording and playback systems for text, voice, and vision. Those systems have been crucial in training the AI systems of today; we will probably need something similar for touch.
There are also very significant safety issues that arise when human-size robots interact with humans, and Brooks provides details on why this is a major problem.
If we are to have robots that can do the things humans do, using the same tools humans use for virtually all tasks currently done by humans (and that is Tesla’s vision), robots have a long way to go. Some good examples of what is missing are listed in a blog post by Benjie Holson.
How about a robot Olympics? Holson proposes an “Olympic Games” for robots, where they must perform a set of 15 tasks. The tasks are easy for any human, but impossible for today’s human-like robots. They are tasks we would want any household robot to be able to do. The tasks are:
- Enter a round-knob push door
- Enter a lever-handle self-closing push door
- Enter a lever-handle self-closing pull door
- Fold an inside-out T-shirt
- Turn a sock inside out
- Hang a men’s dress shirt
- Clean a window with Windex and paper towels
- Spread peanut butter on a piece of bread
- Find the correct key on a keyring and insert it in a lock (without putting the keyring down)
- Roll matched socks (by turning back the open end)
- Open and use a dog poop bag
- Peel an orange (without tools)
- Wet a sponge and wipe a countertop
- Clean peanut butter off your “manipulator” (i.e. hand)
- Use a sponge to wash grease off a pan
Believe it or not, these tasks are beyond the capabilities of today’s human-like robots, even in the lab.
In most of these cases, dexterity limitations and limited touch sensation cause problems. For example, spreading peanut butter on a piece of bread requires a subtle, gentle touch for the hand holding the bread down, but a firm grasp for the hand holding the knife. Similarly, squirting Windex requires a strong effort, but wiping the window afterward requires a soft touch.
Holson notes four limitations that make these tasks hard for robots: no force feedback at the wrists, limited finger control, very limited sense of touch, and “medium” precision of hand movement compared with humans.
How about robots that are not human-like? Brooks makes the case that, for the coming decade at least, robots will be special-purpose machines used for a single task (or a few related tasks). They will more likely move on wheels than walk on legs.
If Musk actually gets the “army of robots” he’s been talking about, they will most likely be rolling, not marching.
On the Kendal Journey 