Google's robot understands spoken English, navigates real spaces, and plans multi-step tasks. It fetches a soda, adapts when items aren't available, and ignores contradictory instructions. The system switches to backup plans when the environment changes.
Questions this video answers
Dr. Zsolnai (Two Minute Papers); content credited to LARRUZ.
October 26, 2022.
"100 instances for $1.10 per hour versus $4.10 per hour with AWS" (claimed 73% savings).
Chapters
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today we are going to give a really hard time to a robot. Look at this. And this. You see, Google's robots are getting smarter and smarter every year. For instance, we talked about this robot assistant where we can say, please help me, I have spilled the Coke. And then it creates a cunning plan. It tries to throw out the Coke can. Well, almost. Then it brings us a sponge. And if we've been reading research papers all day and we feel a little tired, if we tell it, it can bring us a water bottle, hand it to us, and can even bring us an apple. A different robot of theirs has learned to drive itself, also understands English, and can thus take instructions from us and find its way around. And these instructions can be really nasty. Like this one. However, there is one thing that neither of these previous robots can do, but this new one can. Let's see what it is. First, let's ask for a soda again. And once again it comes up with a canning plan. Go to the Coke can, pick it up and Ha ha. Checkmate, little robot. Now comes the coolest part. It realizes that it has failed. And now, change of plans. It says the there is no Coke, but there is an orange soda. Is that okay with us? No, no. We are going to be a little picky here and say no and ask for a lime soda instead. The robot probably thinks, oh goodness, a change of plans again. Let's look for that lime soda. And it is of course really far away to give it a hard time. So let's see what it does. Wow, look at that. It found it in the other end of the room, recognized that this is indeed the soda, picks it up and we are done. So cool. The amenities at the Google headquarters are truly next level. I love it. This was super fun. So you know what, let's try another task. Let's ask it to put the Coke can into the top drawer. Will it succeed? Well, look at that. The human operator cannot wait to to mess with this little robot. And aha, sure enough, that drawer is not opening. So is this a problem? Well, the robot recognized that this was not successful. But now the environment has changed. The pesky human is gone. So it tries again and this time the drawer opens. In goes the Coke can. It holds the drawer with both fingers. And now I just need a gentle little push and bravo. Good job. We tried to confuse this robot by messing up the environment and it did really well. But now get this. What if we mess with its brain instead? How? Well, check this out. Let's ask it to throw away the snack on the counter. It asks which one and to which? Our answer is of course not the apple. No, no, let's mess with it a little more. Our answer is that we changed our mind, so please throw away something on the table instead. Ok, now as it approaches the table, let's try to mess with it again. You know what, never mind. Just finish the previous task instead and it finally goes there and grabs the apple. We tried really hard but we could not mess with this guy. So cool. So a new robot that understands English can move around, make a plan, and most importantly, it can come up with a plan B when Plan A fails. Wow, a little personal assistant. The pace of progress in AI research is nothing short of amazing. What a time to be alive. So what would you use this for? Let me know in the comments below. If you're looking for inexpensive cloud GPUs for AI, Lambda now offers the best prices in the world for GPU Cloud compute. No commitments or negotiation required. Just sign up and launch an instance. And hold on to your papers. Because with Lambda GPU Cloud you can get on demand a 100 instances for $1.10per hour versus $4.10 per hour with AWS. That's 73% savings. Did I mention they also offer persistent storage? So join researchers at organizations like Apple, MIT and Caltech in using Lambda Cloud instances, workstations or servers. Make sure to go to lambdalabs.compapers to sign up for one of their amazing GPU instances today. Thanks for watching and for your generous support and I'll see you next time.
