Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

spectrum.ieee.org

Post reply on HN
Print view View original

11–20 of 70 posts

Re: Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

#11

Post by peterwwillis »

alan_cx wrote: post #8
ryanjshaw wrote: post #2

I always thought this was the only way to build a true AI -- build a 'virtual baby' that has to go through much the same experiences as a human baby. I'm sure this idea has been explored somewhere already - anybody have any pointers?

If that were to happen, I would suggest that an artificial "soul" could or would be created. Once you have that on your hands, things being to get interesting. Ethics, religion law, and so on will have one hell of a job on their hands. Especially when you consider that there would be immediate and obvious military applications.

Do animals have souls, and do we care?

Re: Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

#12

Post by Animats »

This is the bin-picking problem, which has been worked on since the 1980s. For objects of known shape, it's more or less solved.[1] The general case is still a problem. It's good to see Google making progress with this.

[1] https://www.youtube.com/watch?v=TU71MtDC-4E

Re: Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

#13

Post by peterwwillis »

ryanjshaw wrote: post #2

I always thought this was the only way to build a true AI -- build a 'virtual baby' that has to go through much the same experiences as a human baby. I'm sure this idea has been explored somewhere already - anybody have any pointers?

Physical babies take a long time to adapt and form complex connections (http://www.urbanchildinstitute.org/why-0-3/baby-and-brain). There are probably a lot of shortcuts one could take to train it for specific tasks.

In 'The Matrix' they had developed programs that could be uploaded to a physical brain to essentially pre-wire the brain with complex synaptic connections (or so I imagine was the effect). That would be a lot more efficient than waiting 3 years just to get to the point of trimming useless connections. We may have to 'grow' a virtual brain via training to develop the basic platform, but then pre-load it with operations to save time.

Re: Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

#14

Post by duaneb »

ryanjshaw wrote: post #2

I always thought this was the only way to build a true AI -- build a 'virtual baby' that has to go through much the same experiences as a human baby. I'm sure this idea has been explored somewhere already - anybody have any pointers?

> true AI

What does this even mean?

Re: Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

#15

Post by tgflynn »

jing wrote: post #7

Are physics engines not yet accurate enough to enable "virtual" pre-training / full training of the networks, lighting conditions, etc? If they are, exclusively using physical robots seems somewhat inefficient.

That's a very interesting question. My guess is that the physics of grabbing things, especially non-rigid things, is very messy and difficult to simulate. It would be great if someone here were able to give a detailed answer to this question though.

Re: Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

#16

Post by chombier »

tgflynn wrote: post #15
jing wrote: post #7

Are physics engines not yet accurate enough to enable "virtual" pre-training / full training of the networks, lighting conditions, etc? If they are, exclusively using physical robots seems somewhat inefficient.

That's a very interesting question. My guess is that the physics of grabbing things, especially non-rigid things, is very messy and difficult to simulate. It would be great if someone here were able to give a detailed answer to this question though.

It is difficult, but doable assuming Coulomb friction.

The two main issues are that it is computationally expensive, but also that your mechanical modeling has to closely match that of the actual robot (especially the contact model), otherwise I suspect the training data will be useless in the end.

So if you can afford an actual robot, it makes sense to do the training using it.

Re: Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

#17

Post by jackhack »

ryanjshaw wrote: post #2

I always thought this was the only way to build a true AI -- build a 'virtual baby' that has to go through much the same experiences as a human baby. I'm sure this idea has been explored somewhere already - anybody have any pointers?

This is precisely the approach taken at MIT under Prof. Rod Brooks: "Within our group, Marjanovic, Scassellati & Williamson(1996) applied a similar bootstrapping technique to enable the robot* to learn to point to a visual target. " *named Cog, short for "Cognition"

http://people.csail.mit.edu/brooks/papers/CMAA-group.pdf

If I may paraphase, his model is biologically inspired -- believing hierarchical layers of behaviours, lack of a central planning model (distributed processing), and physical and temporal placement in the world (rather than abstractions of the world, or observe/process/react loops) are essential to the formation of a truly intelligent machine.

Re: Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

#18

Post by rm_-rf_slash »

peterwwillis wrote: post #11
alan_cx wrote: post #8

Earlier quoted context omitted.

If that were to happen, I would suggest that an artificial "soul" could or would be created. Once you have that on your hands, things being to get interesting. Ethics, religion law, and so on will have one hell of a job on their hands. Especially when you consider that there would be immediate and obvious military applications.

Do animals have souls, and do we care?

A cow kicked a lantern and burned down Chicago.

A powerful and incensed AI could destroy the foundations of global society before we would even know what's going on.

Re: Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

#19

Post by tgflynn »

chombier wrote: post #16
tgflynn wrote: post #15

Earlier quoted context omitted.

That's a very interesting question. My guess is that the physics of grabbing things, especially non-rigid things, is very messy and difficult to simulate. It would be great if someone here were able to give a detailed answer to this question though.

It is difficult, but doable assuming Coulomb friction. The two main issues are that it is computationally expensive, but also that your mechanical modeling has to closely match that of the actual robot (especially the contact model), otherwise I suspect the training data will be useless in the end. So if you can afford an actual robot, it makes sense to do the training using it.

How computationally expensive ? Are we talking supercomputer time to simulate the few seconds it takes to grab an object ? Advanced robots are expensive too and usually much harder to get access to then computational ressources (ie. AWS).

Re: Google Wants to Solve Robotic Grasping by Letting Robots Learn for Themselves

#20

Post by beambot »

ryanjshaw wrote: post #2

I always thought this was the only way to build a true AI -- build a 'virtual baby' that has to go through much the same experiences as a human baby. I'm sure this idea has been explored somewhere already - anybody have any pointers?

Took a very similar approach back in 2011: http://spectrum.ieee.org/automaton/robotics/artificial-intel...
Post reply on HN