0:00:15 | it's a really long title for a six minute |
---|
0:00:18 | so we caff to convince anybody the robot's are increasingly present in human robot are |
---|
0:00:22 | human environments that sort of open with that one thing i want to address though |
---|
0:00:27 | is that things that we wanna say to robots in those different environments are quite |
---|
0:00:30 | different |
---|
0:00:31 | so the kinds of demands we give to a robot and hospital don't really look |
---|
0:00:34 | anything like the kinds of commands would give to a robot in an office |
---|
0:00:38 | the on that robots have different sensing an activation capabilities different microphones cameras and also |
---|
0:00:43 | like arms maybe norms at all |
---|
0:00:46 | so if we try to develop algorithms that work across a lot of platforms we |
---|
0:00:50 | might end up in a situation where we have to define a lot of in |
---|
0:00:52 | domain data per individual robot platform for individual deployed scenario |
---|
0:00:58 | so my work is focused instead on using dialog |
---|
0:01:01 | as a way to with individual human interactors learn the information of the robot needs |
---|
0:01:06 | on the fly in the particular environment it's displayed in |
---|
0:01:10 | so i'm gonna jump right into the video of that working that will sort of |
---|
0:01:14 | what they're |
---|
0:01:19 | and a sequence |
---|
0:01:21 | no it's not that was four |
---|
0:01:24 | there are so |
---|
0:01:29 | so i give the screen one |
---|
0:01:36 | yes |
---|
0:01:37 | it's never for |
---|
0:01:39 | so as to learn what this they were dreams on the flight figure out what |
---|
0:01:43 | you see |
---|
0:01:49 | jokes or synonym doesn't get one so it actually has to learn the new concept |
---|
0:01:55 | so what it's gonna do is ask me for example rather |
---|
0:02:05 | we're adding one describing policy |
---|
0:02:09 | so there's these objects that are in the room with is anyone's going things somewhere |
---|
0:02:14 | so can ask o |
---|
0:02:17 | a little holes |
---|
0:02:20 | robot a system i appreciate how much that |
---|
0:02:27 | so i show which one |
---|
0:02:29 | the |
---|
0:02:36 | shall not use the word about what is trying to say |
---|
0:02:42 | so negative examples |
---|
0:02:46 | it is played with these architectures |
---|
0:02:49 | so it has feature representation of mileage trying to figure out what the discriminative signal |
---|
0:02:53 | associated with the word rattling is |
---|
0:02:57 | what is right o c |
---|
0:03:01 | two examples is not enough |
---|
0:03:10 | this lunch |
---|
0:03:13 | the |
---|
0:03:15 | so with those three examples it started building like pretty we but reliable classifier |
---|
0:03:24 | so it would require sort of hmm i three five one where |
---|
0:03:33 | you have to trust me that that's the conference room |
---|
0:03:39 | something three five one four three five one |
---|
0:03:47 | yes |
---|
0:03:50 | so now it's |
---|
0:03:51 | i don't be able probably knows something that's gonna go there and find the object |
---|
0:03:58 | space which one best it's you description rattling contain for delivery |
---|
0:04:04 | and again all the objects using this work i have been played with whatever for |
---|
0:04:09 | all shocked looks like we're we show what it does but it's basically like pick |
---|
0:04:12 | them up push them around drop them from all right |
---|
0:04:15 | and for this work modeling units that are learning that picking up an object and |
---|
0:04:20 | dropping it |
---|
0:04:21 | it's a small sound and that's the discriminative signal wouldn't using |
---|
0:04:24 | so these three objects there's like some white something can in the paper container and |
---|
0:04:29 | it decides the paper container is gonna be the rattling object is an example instruct |
---|
0:04:37 | calculating the graph text so on battery power |
---|
0:04:46 | so we find a rattling container and let's go to deliver the box office |
---|
0:04:54 | i kind of regrets feeling of this part where it back so that makes the |
---|
0:04:57 | cute little backup noise |
---|
0:05:05 | something about this |
---|
0:05:08 | all these names but an optimized initialize the system like this with a post relative |
---|
0:05:13 | since all this |
---|
0:05:15 | but it's possible |
---|
0:05:23 | so we do a little hand off should consider |
---|
0:05:30 | so we initialize the system and cannibals are all over that very quickly basically we're |
---|
0:05:35 | gonna and |
---|
0:05:36 | have conversations with humans and ask these questions about local objects to the available classifiers |
---|
0:05:41 | are applied and learn words like travelling in others |
---|
0:05:45 | we're also gonna strength are semantic parsing component |
---|
0:05:47 | by asking questions so when the first and says go the middle lab maybe we've |
---|
0:05:51 | never seen in adjectival construction like that but we do know how to do it |
---|
0:05:54 | with a proposition so it asks where should i go to process the lab in |
---|
0:05:57 | the middle we can now strength our semantic parser by adding this the grammar rule |
---|
0:06:01 | that says like |
---|
0:06:02 | you can say that allow for in the middle and other adjectives work that's the |
---|
0:06:05 | way |
---|
0:06:07 | we test this bunch of tasks about the relocation moving an item from one place |
---|
0:06:11 | to another |
---|
0:06:12 | we quantitatively seen improvement when we retrain both with parsing in perception and we have |
---|
0:06:17 | a user's rate the system is more usable for real world task as what we |
---|
0:06:21 | do both orson perception retraining like this |
---|
0:06:25 | so i think i mostly have time for questions |
---|
0:06:27 | you know that if you have a robot to |
---|
0:06:30 | any system you have a lot more collaborators |
---|
0:06:47 | thank you |
---|
0:06:49 | very quick question |
---|
0:06:52 | how does he noted that dropping an object is a good proxy for shaping to |
---|
0:06:57 | make this time |
---|
0:06:59 | that's a great question so you those examples to try to figure out |
---|
0:07:03 | what we actually do with the low-level is because we have so few labeled examples |
---|
0:07:07 | an expectation for word is build a tiny svms |
---|
0:07:11 | so every svm operates over a particular behaviour and listening context |
---|
0:07:15 | so we have a feature space that says this is what it's like and you |
---|
0:07:19 | listen to your microphone and you draw something this is what it's like when you |
---|
0:07:22 | push down on something and listen to the motors in your arms |
---|
0:07:25 | and then you can use a cross-validation to estimate how reliable each one of those |
---|
0:07:31 | classifiers on the being |
---|
0:07:32 | so in this case like dropping something listening to audio was more reliable than looking |
---|
0:07:36 | at its colour so you are not trusting the classifier for rattle |
---|
0:07:40 | or heavy it's like picking something up and feeling the motors in the arm |
---|
0:07:50 | this is focusing on object actually their performances were given means to those what about |
---|
0:07:56 | the robot itself like maybe doesn't know that doing this is this could possibly learned |
---|
0:08:01 | using |
---|
0:08:03 | no using this framework we have done some work on trying to figure out which |
---|
0:08:07 | behaviors are relevant using like word embeddings just the this kind of exploration |
---|
0:08:11 | but |
---|
0:08:12 | there's like a whole space |
---|
0:08:13 | of trying to do this sort of learning from demonstration |
---|
0:08:16 | and for example where i have become the object and shake it in say like |
---|
0:08:20 | this one models |
---|
0:08:21 | there's something to like when a human watches another human do that we know that |
---|
0:08:25 | the fact that i like should get is actually |
---|
0:08:28 | the discriminative signal |
---|
0:08:30 | and i think there's something therefore like lifting and lowering and shaking |
---|
0:08:34 | in that we can see how someone else does the discrimination and not actually have |
---|
0:08:38 | to |
---|
0:08:39 | do this svm estimation |
---|