[00:00:05]
>> He did his bachelor's degree in mechanical engineering at the University of Dayton event is masters of his Ph d. at mit and then Ivy did his research alongside Neville Hogan and his work was done in the comics who were designing control actuators so that they could interact stable and efficiently with the environment and this was some really important and groundbreaking work that you still see the influence of that work and influencing a lot of Mobile Robots and that included in fact and I think the Sandia wander ovata hopefully you'll see a little bit of you even see it in the form of the novel who is transmission designs and you can see it here on campus where people are incorporating some of the stuff that Steve was one of the 1st people to explore so I think that's really cool this through the back and Steve is really an ideal iron's 1st of all as I just mentioned he leads a lot of really innovative and exciting scientific research Erith at Sandia 2nd he takes those interesting and exciting ideas and puts them on to feel Vol I consequence of platforms that really make an important impact for national security and for those of you guys who do our you know how hard it is to take research and put it on platforms that have to work every time and finally it seems just been awesome and I've had the pleasure of working alongside him and I've seen how he can mentor students from high school students to undergraduates to host doctors grad students and I think this talk a little straight out he's capable of leading everybody's Everest team getting everyone in Dave's an excited about the work and therefore I think it's going to be really great top Thank you all for coming and thanks Kate.

[00:01:42]
Well thank you thank you thank you thanks for that kind introduction and I think that's the introduction that I wrote Ok I appreciate I appreciate the kind words and I thank Professor Hutchinson for the invitation c.j. for all the logistics here in and for initiating my visit here and all of you for taking some time over lunch hour to 2 to hear about what we do is today for my own interest and benefit can I just get a quick show of hands a who is familiar with Sandy or feels they know what we do Ok that's maybe half a little more that's that's not too bad that's that's good I didn't know much about Sandia before some folks that I worked at with mit started getting jobs out there but we're for those who don't know where a federal research laboratory New Mexico is on imagine we work with a Department of Energy primarily in our sponsored agency is a National Nuclear Security Administration though by this chart you see that about half of the work we do the last 30 or so years is for other entities almost all federally funded but we work sort of across the federal spectrum and do a bunch of different things it's a good sized place most of our people in Mexico we have a substantial site also in California and a few other satellite sites that are smaller.

[00:03:02]
My group is the consequence automation robotics group so we in a kind of sort of a slide for the decade and a half or so that I've been there live almost entirely on the right hand side in the non n.s.a. world doing a bunch of activities for a bunch of different customers across the technology spectrum we also do work that ranges from research probably in a university context more of what you would consider applied research development and then we field real systems across these different areas my team in particular which is basically originated almost all the work that I present today is about 5 or 10 people out of the 50 or so in this broader group and we sort of are on the most research the end of what we what the group does at Sandia So we'll we'll touch on a few things this presentation going to be a little bit more broad than deep Thank you.

[00:03:55]
In order to touch on some topics so feel free if there is curiosity we can go deeper into some of these topics I'll do my best to answer those questions if I can't then I'll just slowly walk out the door and let any. Students best. So. I will talk about a few different topics.

[00:04:15]
For me but the one common theme that will run through these is it's kind of a future use of robotic and I mean systems the my team sort of has its eye on which is for lack of a better word sort of tactical application so trying to bring robots on and systems into environments that are extremely challenging because they're the goals may be dynamic means to have some sort of high level goals but the environment itself is changing as you operate so these operations can't be scripted we need to do things quickly there are potential adversaries trying to thwart what you're doing depend on the application or it's just very challenging so these are cases where we're trying to.

[00:04:53]
Reduce exposure of people to really really difficult and challenging situations or leverage and augment their capability to do that we need to be about as good as people and about as fast as people are also won't be interest in doing it and this is really hard and I won't try to claim that we can do this but this is sort of the long term goal that we're working toward which requires a bunch of pieces so just a few sort of motivating examples that we think about this could be a case where there's some facilities it's unknown it may have threats we want to bring our sort of robot team to go check it out and see if there's trouble there there might be kind of weird in turns dressed all in black lurking around.

[00:05:33]
We want to find. Another another area would be things like mine rescue or disaster response to a potentially very challenging terrain this pictures from something from Fukushima we have just absolutely insane terrain very difficult to handle in cases like mine rescue you have very difficult communications environments necessitating us of a high degree of autonomy so conducting operations and again life may be at stake so you have to move quickly another example this is a graphic from DARPA but future concepts of operations would involve sort of human robot teams moving together at the same pace again dealing with uncertain.

[00:06:09]
In this case the soldiers will be directing what they want the robots to do but. They need an intuitive way to get the row us had to do what they want. And finally an application that we care a lot about in the broader organization which I work which is physical security so you might have this facility that is high value need to take care of it you don't know where the threats are currently it's very expensive to build these systems in the just throw a whole bunch of sensors at it they get a whole bunch of alarms they don't really know what to do so we want to sort of understand what's going here want to bring robots to bear on this and understand sort of the adversarial environment so this requires a bunch of pieces I want to talk about all of them and talk about some sometimes mobility is very important energy efficiency in order to actually sustain operations is not good enough to just be able to move you have to be able to actually operate for a while you got to go fast sometimes it's collaboration multi robot collaboration or collaboration with humans perceptions very important tactics those kind of the autonomy pieces and then actually in many cases taking action once you actually get somewhere what do you do to make a difference so.

[00:07:13]
Think about the autonomy piece of this challenge I like to use this kind of letter sure we won't talk to too much but at the bottom this is basically your operating something with a joystick and as you move up you're getting more and more capability from the robot without any operator intervention there a few key technologies that have emerged over the last few decades and mature that help the cause a great deal so for example the maturation of some attendees localization of mapping algorithms enables robots to at least in research applications so far develop very rich maps potentially with very attractive human displays of unknown environments rescue center in a building come out with a map the thing to realize though may be obvious but it's important enough that the robot to the robot this is really just stuff vs not stuff for us the robot has no conception of what the world is that's really important because it gives the robot the ability to navigate around and get back out but it doesn't actually know what it's looking at so if we want to get to the sort of higher level sorts of behaviors we say want to go into a building we don't want to map that we want to find all the gas tanks and swap them all or something like that we need to understand the world in a more abstract level so we really see sort of this perception piece as the key enabler to anything else that might come further up the chain and we've been focusing on that so now we want to develop sort of maps of the world where we also label objects and we do that in real time and relate those object labels to the 3 d. geometry that we're producing as we go along.

[00:08:40]
So the title of my talk includes term active perception and I want to talk a little about what I mean there because I think there are really 2 pieces of this problem that we need to think about how they how they're each implemented and how they tie together so on one hand you want to take all the sensor data you get and draw some conclusions about what you're observing So you want to do this sort of semantic classification that we talked about so.

[00:09:10]
So if you're in something like our prototyping to back it back at Sandia you'll be able to say Ok there's cabinets there's a mill Lay there's a bandsaw whatever and I also want to be able to say Ok this looks like a machine shop it's got those items you know might be might want to say you know there's a there's an there's a volume back there that has a door going to it that might be another space so might be people back there that might be something of interest so either level of semantic inference might be of interest the flipside of the problem is that I actually want to figure out what data I want to get it so the idea here is what I really want to do is drive.