Robots That Jump

Robot Bodies Needed Before Robot Minds

Monthly Archives: September 2014

Robots that Shuffle

Progress on the development of agile robots, with the Boston Dynamics/DARPA humanoid robot Atlas dragging a heavy object. In particular, the heavy truss being dragged is “unmodeled” – in other words, the experiments didn’t fix the results by pre-coding the dimensions and weight of the object into Atlas. Dynamic adjustment = one small step for Robots that Jump!

https://www.youtube.com/watch?v=oYSitvLW0r0

Compare to related videos by the Atlas group:

https://www.youtube.com/channel/UCrnmsv8fmwVJpQs0YFDN7rg

While this demonstration is impressive (given the incredibly lame performance of humanoids over the last 20 years) it still is far from a “robot that jumps.” In particular, the robot had internal sensors but no external sense. The slow, old-man shuffle the robot displays prevents the truss from banging into its legs. At the speed the robot is moving, you can see the truss tapping the lower leg on a regular basis.

While animals may be clumsy, they are not clumsy in this way. The reason is that they are “sensor-first” and “sensor-dense.” Classic robotics emphasized exhaustive computation from a few sensors, rather than shallow processing from a very large number of sensors found in living things. The result is robots that can do old-man tasks, but have trouble displaying the dexterity often touted for our so-called replacements.

Where will this work go? Given the interesting work Boston Dynamics did on multi-legged robots, one has some hope for Atlas. But it runs the risk of being another grad student demo, despite the fact that the methods of coding Atlas responses seem closer to biology than is typical in these projects.

But, as usual, this is lost on the tech industry, whose “reporters” still pad their stories by implying the robot is much more than it is…

It’s a good thing robotics engineers haven’t figured out how to make humanoids disgruntled yet.

Gizmodo (of course)

Moving smoothly in the world is a arthropod-level function. There are about 50 more levels necessary on any robot (even if Atlas moves faster than a super-old man someday) before we get to “disgruntled.” And it is NOT scary to see a machine do this – instead, it is a bit pathetic. Our machine overlords turn out be be shuffly old men.

Why has implying robots can do more than they can turned into a cute way to end a tech story? This is not the same as saying robots may be dexterous someday. And why do we need to be “scared?”

Frankly, I’m more frightened of super-sized cellphones.

Advertisements