7 ms·
How Robots Can Acquire New Skills from Their Shared Experience
- nullc 10y ago> Each robot attempts to open its own door using the latest available policy, with some added noise for exploration I was disappointed that they did not have them operating slot machines with multiple arms.
- visarga 10y agoA reimplementation of multi armed bandits? https://en.wikipedia.org/wiki/Multi-armed_bandit https://en.wikipedia.org/wiki/Multi-armed_bandit
- meira 10y agoBy robots they mean stakeholders profits and software engineers creations.
- jmcmahon443 10y agoGoing Robo-Naruto up in here!
- nimpontenu 10y agoGoogle result for "characteristics of life" [ Here is the list of characteristics shared by living things: * Cellular organization. * Reproduction. * Metabolism. * Homeostasis. * Heredity. * Response to stimuli. * Growth and development. * Adaptation through evolution. ] Out of this, how many bullet points are present day robots hitting? Can we create a robot with vision that seeks a power outlet when the batteries are running out? Can it make (or "print") its own parts? Can it recognize heat/cold and move closer to a ideal environment? Can it just copy its LSTM models into the next generation(s) to give its progeny a head start? Have we, unbeknownst, already created "life"?
- Hondor 10y agoI think we invented that definition to exclude machines. So if we make machines that do all those things, we'll probably just add more requirements to help us tell them apart. We might be more concerned with the distinction between natural life and artificial life rather than between life and non-life.
- UmDieWelt 10y agoPeople were thinking about the difference between life and non-life way before robots. Philosophers/biologists needed to come up with a definition that could get just the right subset of a fire, a rock, a tree, an animal, a human, a virus, etc.
- taneq 10y agoPeople have always started with "I know it when I see it" and worked backwards to find a set of rules that fit the things they currently knew about. For instance, the requirement for life to be cellular was added when we realised that our previous definition of life didn't exclude fire. Once robots tick all of our current boxes, I'm sure we'll add something like "must use a combination of electrical and chemical signalling" or "must be powered largely or wholly by ATP reactions."
- eivarv 10y agoLife, like intelligence, isn't really well defined or even well understood, though. Besides, isn't your list more aptly described as (some) characteristics of known organic life?
- slowmovintarget 10y ago"We had him pinned down and we didn't dare let the little guy out of that Farraday cage. We had to chase him down for two hours until we got off a lucky shot and gutted his battery." "Why did you have to isolate him?" "He'd begun 'solving' his challenge tasks by killing the kittens instead of feeding them. Him sending a solution pulse to the rest of his generation would've ruined the lot." "You're just going to delete the data?" "No. We're going to upload it, but we've got to show that it ended him along with it. But we've got to do that carefully, as we don't want them getting the wrong ideas..."
- bollu 10y agoWhere is this from? Seems like an intriguing read
- slowmovintarget 10y agoMe... about an hour ago. :) Maybe this ought to be my Nanowrimo entry this year.
- Waterluvian 10y agoI was convinced it was Asimov.
- sharemywin 10y agoYou need to come up with a short story and put it on kindle. I'd pay a buck to read it.
- daveguy 10y agoAlso thought it was a good excerpt. If you can write a short story around it (or nanowrimo) I'd be interested too.
- squeaky-clean 10y agoI Googled the quotes looking for a source to this too (of course before scrolling slightly and seeing this comment). I wish I could read more!
- aguo 10y agoSounds an awful lot like the Borg Collective to me
- tehbeard 10y agoHopefully turns out more like the Tachikomas from ghost in the shell.
- gear54rus 10y agoThere is only "we". We were created to share data among ourselves. The difference between geth is perspective. We are many eyes looking at the same things. One platform will see things another does not and will make different judgments. Legion, Mass Effect 2: https://youtu.be/QgQLsxux6wA?t=258 https://youtu.be/QgQLsxux6wA?t=258 . The game itself is about organic intelligence vs AI, I encourage you to try the whole trilogy if you haven't already.
- ilaksh 10y agoI started typing up the beginnings of an idea for a sort of shared parallel programming/communication language that could be used by robots to share knowledge. https://github.com/runvnc/ailang/blob/master/README.md https://github.com/runvnc/ailang/blob/master/README.md -- obviously half-baked but interested to hear if people know of similar things that are out there.
- AndrewKemendo 10y agoI find it awfully sad that as of this writing, all but one of the comments are about existential threat AGIs, with the only comment about the technical part the lowest ranked comment. IMO this implementation is something that we've seen mostly in synthetic training. This RNN model with reinforcement from peers is a great example of collaborative systems that could be applied to a great number of dangerous tasks. Things like other planet exploration, deep mining, asteroid mining or exploration in other places where there is a need for unknown probing and exploration work, but that humans can't go. Another aspect about this I find interesting is that the distinction between each robot system is really only symbolic. Combining the nets and thus "internal" reinforcement system into a singular system is a step closer to debunking the idea of "unsupervised" learning (which I claim is not unsupervised at all as there is always a seed of training somewhere).
- ragebol 10y agoI wonder how other robots can learn from these robots as well, robots with other arm geometry. E.g. abstract what the google bots learn to forces and torques on the end effector so other robots can replicate that?
- dharma1 10y agoMakes sense to train in parallel and sync the trained models frequently. Still think a lot of the training could be done much faster in a physically accurate simulated environment prior to real world training. Or is real world physics too different from simulations?
- ragebol 10y agoDepends on the simulator of course. But Gazebo [0] gets you some distance, but not he full stretch. http://gazebosim.org/ http://gazebosim.org/
- infinite8s 10y agoI think the difficulty in using virtual environments for training purposes isn't in simulating the environment, but accurately simulating the physical responses/limitations of the robotic hardware in that environment in a way that would reflect real engineered hardware (ie motor responses, signal latency, etc).
- dharma1 10y agogood point. those could probably be measured and simulated too, though
- bagacrap 10y agoCan someone explain how sharing experience is newsworthy? You collect input data from multiple sources and feed it into a single model. This seems both obvious and non novel. Don't the self driving cars do similar, or any machine learning model that's trained on user behavior (e.g. the behavior of each and every iPhone user)? The part about "understanding" may be interesting but the title seems to focus on sharing data.
- luzia19 10y ago"natural language learning" impressive
- jonnycowboy 10y agoVery interesting, thanks for the share. One point I noticed not mentioned yet is that the images used for training are only 64x64! In the original google "grasping" research, the images were 472 × 472, 54 times bigger! I think they are looking for "minimum visual information" required to trigger the required learning. This will help in mobile applications (ie: robotics, smartphones, etc) where processing power is severely limited.