Daniel Lofaro talks about how he works hard to make things easier and how co-robotics are the way forward, and how bringing cost down cost and better AI is critical for this. He talks about how robotics needs a “killer app”, something that makes it compelling enough for the customer to take the step of welcoming a robot into the business or home. Daniel also discusses creating an ecosystem of robots and apps, and how competitions can help do this.
Daniel Lofaro’s PhD project made its public debut on an unlikely stage. His research in end-effector velocity control of non-planted robots (robotic throwing) enabled his HUBO humanoid robot subject to throw out the first pitch of the 2012 Major League Baseball season in front of 45,186 fans.
Daniel is Assistant Professor in the Department of Electrical and Computer Engineering at George Mason University, and the director of the laboratory Lofaro Labs Robotics which is apart of the international laboratory group called the DASL Autonomous Systems Lab Group (DASL Group).
Additionally, he's a affiliate faculty at the U.S. Naval Research Laboratory (NRL) in the Navy Center for Applied Research in Artificial Intelligence (NCARAI) within the Laboratory for Autonomous Systems Research (LASR). As a NSF-EAPSI and ONR-SFRP Fellow, he received his doctorate from the ECE Department at Drexel University in 2013 under the guidance of Dr. Paul Oh. He was the Research Lead of the DARPA Robotics Challenge team DRC-Hubo from 2012 to 2014.
Host: Per Sjöborg, Robots in Depth is supported by Aptomica.
His research focus is in the overarching field of real-world robotics. Within this his research interests include Swarm Robotics, Emergent Behaviors, Robot Design, Real-World Human/Robot Interaction, Humanoid Robotics, Complex Control Systems, Secure Robotics, Cloud Robotics, Unique Musical Instrument Design, and Real-Time Systems. Research interests include Complex Control Systems and Robotics with most recent ventures relating to Robot Design and Cloud Robotics. Daniel's dissertation title is Unified Algorithmic Framework for High Degree of Freedom Complex Systems and Humanoid Robots.
This podcast is part of the Wevolver network. Wevolver is a platform & community providing engineers informative content to help them innovate.
Learn more at Wevolver.com
Per Sjoborg: Welcome to the podcast version of Robots in Depth Episode 10 with Daniel Lofaro in cooperation with Wevolver. Today, I'm honored to have Daniel Lofaro here and we're going to talk about everything in robotics like we usually do. We’re going to start the regular way. How did you get into robotics? When did you discover you could build things that move around in the real world rather than just code?
Daniel: Why robotics?
Per: Yes, why robotics?
Daniel: My question to everyone is why not robotics?
Per: That's also my question.
Daniel: Robots are awesome. There’s no reason why anyone shouldn't be obsessed with them. I go by a motto that I work very hard to make my life easier and robots are a way to do that.
Per: That's also why I'm into robotics. I love all the stuff you can do with computer coding. For this project I've coded away all the boring stuff but I'm envisioning a world where you can do the same thing for physical tasks like we're now doing with cutting our grass. If you have a big lawn you have an automated lawnmower and it just works a treat.
Daniel: In co-robotics that's the future. For those who don't know what that is, it's connecting, having people be able to work next to and work with robots instead of in the normal industrial setting where you have a robot inside of a big cage or behind this big yellow line that you're not allowed to cross otherwise you'll be cut in half because in general these industrial robots are very stiff. They’re very strong and us humans were very squishy. That means that we are going to get hurt if the robot hits us and the robots not even going to notice that it hit us.
Per: It doesn't have the perception to go in and say okay, there's a human. It’s got a very simple if you cross the line it stops. It’s not intelligent in any way.
Daniel: It's not intelligent and if you cross the line it won't stop. It’s up to us to protect ourselves. With a co-robot that's something that we could work next to. It can hand us tools. It could screw in something while we're doing something else.
Per: They could hold a big heavy piece while we do the screwing.
Daniel: Exactly and so their actuators are made to be more elastic, softer so when you push it it'll give. Just a human if you push them they're not going to stay perfectly rigid like a robot traditionally would and this will allow them to actually work next to us to share a common space with humans. That’s exciting.
Per: That's the way we interact with most machines today. I mean you don't fear your coffee or your stove or your washing machine or your car. I mean the car must be the ultimate co-machine. We are actually in the machine and then constantly operating the machine as we do. This is by far the largest part of the market and it opens up so many different fields. It’s going to make everybody feel like a superhero because they can do these things that we today are not even collectively able to do. What do you see as the major challenges to get there?
Daniel: There's two big things, first of which is free and the other one is very expensive. Free, I'm sorry all you programmers out there but software should be mainly open-source and it's your artificial intelligence, AI. That needs to improve dramatically so we that can actually do tasks that we want them to do so they can deal with our complex and ever-changing world. AI is something that technically doesn't cost anything to reproduce and is a big game changer. Big companies Google, Amazon, Uber all know that this is a game changer. They’ve put a lot of money into AI right now.
Per: All of them and I guess both three-letter agencies and research labs are also on this.
Daniel: DOD, DARPA those places. NSF is big into it as well but the second thing which not everyone thinks about is cost. We need to start mass producing these items. We can't have a robot such as the wonderful HUBO series which cost over $400,000 apiece. What can it do? It can walk around and it can fall. I love that robot but it's $400,000 for something that can't do much of anything in our everyday lives. If that was $10,000 or $20,000, the price of an automobile and it can do your dishes, it can make you dinner, it can vacuum your floor and things like that and last 20,000 steps or a 100,000 steps just like you expect an automobile would last a 100,000 miles. $20,000 will last you 10 years great, that's acceptable. Right now since many of these things are one-offs or for the example of HUBO, we have about 16 HUBO II pluses out there and I guess about eight DRC HUBOs. Those are small numbers but if you start making these things in quantity just like cars. Custom cars are very expensive but buy a Honda Civic for $20,000 or less and those are good cars. They’re just done in quantity.
Per: In one way they actually get better by being made in quantity because we all know that your favorite super sports car is going to spend a lot of time in the workshop but your Honda Civic is just going to run for 20 years because they have different focus. You also as a programmer and as a developer you also know this that the more input you get, the more feedback you get, the more bugs you can get out of it and that is also, we are both the cost thing, if you do more of them they're going to be cheaper but we're also going to make them better by having more testers out there that's going to call us up and scream at us it doesn’t work and then we can fix it.
Daniel: That is a great point. What we need for robots is a killer app. I don't mean that the robot is going to murder us. I mean that it's the next Angry Birds. The iPhone, I love my iPhone. I use Apple although I am switching to Android for various reasons. I started using the iPhone because someone showed me how to play Angry Birds. I'm sorry. It’s just an obsessive game and I think a lot of people are that way. I'm sure the video conferencing is good. You get to have your calendar but I have games in my pocket all the time and that is great. It fills those voids of time that we don't have anything else to do. Robots, we need something for robots that are as obsessive, that are as awesome and as useful for the robots. If there's a lot of them out there, they use a ubiquitous framework then you have a large user base and they can program apps such as a kitchen cleaning app, a vacuuming app and people can make their own. Upload them to the robots, maybe sell even sell them on the robot store and that would be the future because now you're crowdsourcing applications. It’s not Apple that puts out all these games.
Per: No, Apple didn't do Angry Birds.
Daniel: Rovio did. Rovio made Angry Birds but it made Apple more money than it made Rovio.
Per: It's having this ecosystem, getting it out there and cost and also features, it's a very big problem for robotics today. They’re very expensive for what you get back and that is a huge challenge. How do we solve the chicken and egg for the robotics world? How do we get the price of the hardware down? Before that nobody's going to develop an app for a robot there are only eight people have access to but nobody's going to develop a hardware where there is no, how do we how do we kick-start this? How do we get this going?
Daniel: That's a good question and I think it does require a bigger company to do something. In this case it wasn't a company. This is DARPA, the Defense Advanced Research Agency. What they've done in the past with the DARPA Grand Challenge with the autonomous vehicles, they said they see a future for autonomous systems in our everyday lives i.e. autonomous vehicles. All right, they wanted you to drive 170 miles through the desert autonomously. They did it twice the first year. No one succeeded. The second year, multiple teams succeeded. Let’s move to the urban challenge. Now they did the same thing inside of an urban setting. The robots were able to read the road signs. They were able to stop. They were able to go, obey traffic, avoid pedestrians, avoid other cars. Then what the government do? Nothing. They handed it off to industry because then they could see that there is use for. Now what are we seeing? Google cars, autonomous vehicle. Tesla is now autonomous.
Per: Actually handling I would say from a time perspective for most people stop and go in traffic and highway is 90% of what you drive.
Daniel: The key point here is through DARPA, government agency saw a need for autonomous systems in our everyday lives in order to better our lives. It wasn't cost beneficial for companies yet to put up the money so as what our government should do they want to help make our lives better and they put up the money initially to get excitement about it to show that it is feasible. They did it and now the bigger companies have gotten a hold of it and they're doing great things with it. With the DARPA Robotics Challenge, the most recent one from 2012 to 2015, they do the same thing with humanoids. Adult size humanoid robot be able to drive a car, open a door, turn valves, go up steps, go over rough debris, use human tools, etc. just to push that boundary to try and
Get companies a little bit more excited about these things. We saw Google after this started they purchased a bunch of robotics companies including Boston Dynamics. It’s very exciting to see that even privately owned companies are getting excited about robots at the moment. Amazon excited about UAVs. That's one thing but they use robots in almost everything in their company from moving around their supplies to now they even have the Amazon Picking Challenge where they want to have robots actually pick the items out of the shelves that the robots carry to them. It’s really exciting that again the bigger companies are getting involved now and that stems from DARPA, again the DARPA Robotics Challenge.
Per: DARPA and their previous ARPA was course the father of the internet to a large degree too. They’ve done this for quite a number of times and they can be successful on a scale that's hard to believe in many times.
Daniel: Not even to a larger scale ARPA, it's what DARPA used to be as you stated was the father of the internet. The first connected computer was I believe in 1969 and it was called ARPANET. That was the birth of what will eventually become the internet.
Per: Where do you think we are in this spectrum? Are we 85, 95, 2005?
Daniel: I'd say we're in the very early 80s or late 70s. The reason for this is one of the big reasons why the internet exploded, I mean again we had the internet for over a decade before AOL had its AIM, instant messenger. Why was AIM so popular? Because everyone had it. It was ubiquitous. Why did websites become popular? Because they all use the same protocol HTTP. Things standardized. We started to be able to use one thing on different hardware so everyone view it. That’s what made the internet boom making things run on other things. Robots are not there yet. It is difficult to run the same software on the same series of robots let alone a totally different kinematic configuration. We need a standardization. That’s why I think we're even before the internet as we know it or even the earlier internet and as we know it today. We need to standardize.
Per: Yes, I fully agree with you. Standards are absolutely essential. I'm also a fan of open source. I am loving the ROS, robot operating system. What’s your point of view on ROS? Where are they now?
Daniel: Let me start off with saying that yes, you guys have done a great job. You put a community together and that's the big thing about ROS is people put a lot of time and effort into adding functionality. ROS itself, I apologize is far from a good system. Every build breaks things in the previous build. This is a huge problem. You can't run something that ran on an older version with the newer version. It just can't do it. It doesn't run in real time which is what we need for robotics. I'm talking about high frequency, real time guaranteed delivery. That’s purely because of how it was originally designed. It does what it needs to do but those were not its focus. They do have ROS 2.0 that they're working on which is hitting on these subjects I speak of but hasn't been released yet. There’s a beta version out.
Per: They're also aware of these issues.
Daniel: They're very aware. I've been known to not be a fan of ROS that's because I'm big into real time systems. ROS is excessively far from a real-time system.
Per: With this 2.0 effort that I've heard about that is coming out is that raising your hope in any way?
Daniel: Yes so 2.0, it's exciting so what they say that they're working on is they're putting stuff working on real time guarantees, package receive all guarantees as well as making it lighter weight. ROS when you install the full package goes over 3GB. Why? Now they're also trying to make it run on much less powerful systems such as these single board computers like Raspberry PI's or Android phones etc. that way you can run on Internet of Things which is the new hot topic out there. That gives me hope.
Per: Have they told anything about when this is going to be available and what what's your take on that?
Daniel: I believe if memory serves me correctly that they were supposed to already have released the version but as with everything things get slow. That's fine but remember this is an open source movement. They’re not necessarily getting paid to do this. We would need something again from DARPA to make it better. Again with the DRC, there was a, back in 2011 there was a few major simulators for robots. Two of which Open Rave, the other one was Gazebo. They were both fine and they both had their problems. The DARPA Robotics Challenge adopted Gazebo for their simulator. With that came a lot of money to put in real development into it, still open-source. Now Gazebo is far superior to Open Rave and is a quality system that people can get up and running very quickly and get good data out of and get good simulation out of. That was with only a little bit of money and a little bit of effort for the DARPA Robotics Challenge. I think what needs to happen is another initiative to standardize these robotic platforms, find out what's needed and build something because we can do it. We can't do it as a side project.
Per: I think the ROS community is already contributing as much as they can. Somebody else has to come in from the side and realize that this is very and this has to be done and as you say, if you compare it to many of the other government program it's going to be very little money that's going to return so much to society. We really can hope we can encourage somebody to take care of this.
Daniel: Again to reiterate the ROS community, that's what makes ROS good. Most people who use ROS they will complain about how ROS doesn't connect, it doesn't transmit data well, etc. but they use this, this and this package and it works. It’s not something that the Open Source Robotics Foundation made. They’re just using their IPC, their Inter Process Communication to use a package someone else made. It’s kind of like almost like the printing press for robotics. You’re able to make an app, send it out to other people, now they can use it just like you're able to write a story or an article, print it out and share your ideas with other people. The same thing with Thingiverse and 3D printers. You can have an idea. You can upload it and suddenly it can materialize somewhere else. Doesn’t that sound like something interesting? Star Trek right? A replicator.
Per: The Star Trek world is amazing.
Daniel: It's not quite make me a cup of Earl Grey, hot but you can say make me a cup and it'll do that. We just need to work on the Earl Grey part.
Per: It is an amazing time and I really hope that we can keep the operating system, the software and even keep the standards open-source, doesn't mean the hardware have to be open sourced but the standards for them connectors, control systems, schematics and stuff so that everybody can build stuff to add to it or add-on it or build with it. Do you see any alternative to ROS or we have to save it because that's the only way to do it or do you see any other roads ahead?
Daniel: I think there are other roads ahead. There’s again ROS 2.0. I hate to toot my own horn but you know act based systems which is for example HUBO Act which is a system that can talk to either remote or local processes just like ROS can but in high speed, real time, very little latency which is ideal for things like humanoid which what I work on. Barring those there's no great unified framework and that needs to happen.
Per: I mean you're aware of this and I presume others to be, do you see a movement towards this solution for this problem? Are people talking about it?
Daniel: There's a problem. I do not see anyone moving towards, everyone does their own in-house version. Russ Tedrake up at MIT, they have an absolutely wonderful real-time UDP connection between MATLAB and their robots so they get really low latency. You get kilohertz speeds. That’s their solution. Works on some things but then Georgia Tech can have their own, Purdue is going to have their own, George Mason is going to have their own. It’s faster to do an in-house thing and just get it done for your real research as opposed to do something overall.
Per: Because you have to talk to a lot of people and coordinate a lot of stuff. It’s faster because you can go straight and you don't have to do something that's general which takes longer time. You can do something that suits your particular needs case.
Daniel: What there needs to be is a robot operating system. I am not talking about ROS because that is what ROS stands for. ROS is not an operating system. It is an inter process communication, it's a communications framework. An operating system is something that stands directly between the hardware and the software. We need to have like an Ubuntu for robots where you can install it on any machine and it works. You can open up a browser. You can open up OpenOffice and you can now surf the web. You can now type a document. We need to have something where it has all the common drivers for robots, for all the common robots becoming inverse kinematics for these robots, common ways to do path planning. You can say robot go over there and it might not use the most efficient and the latest and greatest method but it'll do a method that works pretty darn well.
Per: And is also proven to work.
Daniel: The out-of-the-box Ubuntu is not the best. People usually tweak it to their own needs or there's are things like Gentoo which is really tweaked to their own needs but most people are perfectly fine with it just works well enough and that's what needs to be made but there's no funding for that at the moment.
Per: That's something we have to work on absolutely.
Daniel: We have people full-time to just put it out. Once there is a base system that has the major things what has happened people contribute because now they want their robot thing in there. I need this one in there. I just made this new general case off. We’re going to put that in there too. That’s what needs to happen.
Per: Also companies will start develop additions and products around this like we've seen for the WordPress community you can get a plug-in for four things you didn't know you needed.
Daniel: Let’s look at the video game industry. Let’s go back to the 80s.
Per: It wasn't much in the form of video gaming at that time.
Daniel: There’s plenty of video gaming in the 80s and 90s. It was all on either a console or Windows because that was the one that was the programmers programmed on, most people had them. It was just a large market share. Now, it's not quite 90% Windows anymore. A large portion have either a Mac or a Linux computer. What has happened? Steam has happened which is something that can run on multiple different platforms including Linux so now you can have a Linux computer and still play Unreal Tournament because that is what people want to do. The support for other operating systems that aren't just Windows is increasing and increasing.
Per: Do you see any hardware issues we need to work on? We’ve talked about the software and standards and stuff but is there a sensor type or is there an actuator type you think. We really need this because otherwise it's going to be hard to make them work.
Daniel: The biggest thing that's going to change robotics is materials. Absolute biggest thing because we already have some really good tech out there but you make it stronger and lighter you just made it better.
Per: A cell phone weighing a kilo isn't that interesting. One weighing 100 grams is very interesting.
Daniel: You start making stronger alloys that are also a lot lighter so they can change your harmonic drive which is steel with something that is some other alloy that is significantly lighter. You don't need as much power to move your robot which means you don't need as large of a battery or let's say you make a more efficient battery. Right now the HUBO series robot she'll stand around for two hours but she'll move around for less than 45 minutes. That’s not a full day. We wouldn't use our cell phones if they only lasted for 45 minutes.
Per: If we look at robotics in total and we see where are you most optimistic about the general public actually being able to use a robot. I guess it's not a humanoid because those are complicated devices compared to a vacuum cleaner. What’s the next vacuum cleaner so to speak?
Daniel: I would actually argue that the thing that's going to make robotics ubiquitous with everyone is a humanoid robot.
Per: Because of its generality it can be applied in many situations.
Daniel: Exactly so humanoid robots are the Swiss Army knives of robotics. They can do everything. They don't do everything well. A humanoid robot can pick up a box. A crane can pick up a box that's a lot heavier so they can do a better job but when was the last time you saw a crane driving a Honda Civic down the highway? A humanoid robot can do that as well. It can make you a sandwich. It can vacuum your floor. They are very useful.
Per: The same robot is useful for many different things. You don't have to make custom ones that make them expensive.
Daniel: You don't have one car that takes you to school. You don't have one car that takes you to the grocery store. You don't have one car that takes you to the park. You don't have one car that takes you into the city. That’s silly. Why would you even think about that? Oh you know because city roads are different than the park roads, oh we have to go on gravel here. No, you have one car for all of this. Why would you have more than one robot for all of this? You already have the vacuum cleaner and that's why it has to be human shape because form follows function. Our world is made for humans. It has doors that open. It has light switches you switch. It’s made for things that has two arms, two legs and a head. In fact most robots should be right-handed because our world is made for right-handed people. 90% of the human population is right-handed.
Per: Very interesting and why I like the humanoids are the generality. I mean they can potentially and theoretically do anything a human can do. Do you see them coming down from the $400,000?
Daniel: Yes, they're definitely going to get cheaper. In 2008 Jaemi HUBO, one of the best humanoid robots ever. She’s now down at a UNLV, at our partner lab at UNLV. She costs $1 million in 2008. It’s 2016 now, eight years later you can buy a superior robot to Jamie and approximately the same size, weight, etc. for under $400,000. Dropped over half in cost and this is partially because at the time when Jamie was made that's a KHR4 model. There was Jaemi and Giyopo so two of them that's it. Now the two pluses which are the superior versions but are almost identical to the other ones minus hardware both electrical and mechanical upgrades they were made in quantity. There’s about 16 of those out there. They’re sold to the US, Singapore, places in Korea, China so because even at a small quantity like that they were able to half the price. That’s exciting.
Per: Halfing the price kind of adds a zero to your market potential market. If you go from two, they might even come up to a hundred in the long run and the next one adds, it goes up to a thousand when it comes to $40,000 or $100,000.
Daniel: Hardware can be fixed with economies of scale. Mark Tilden can talk to you about that. You make one widget. It might cost you $50. You can make 50,000 widgets it can cost you $1 each so economies of scale. That’s the same with hardware but economies of scale is not going to fix the artificial intelligence. This is where the robotic app store comes into play. Again, you're saying we don't need the artist. Oh maybe we do, maybe we don't. We’re not asking for the general thinking robot. I'm asking for a robot that can do my dishes. Download the app for that. Yes, I will pay $20 for an app for that. I need a robot to vacuum my floor. Yes, I will pay $20 for an app for that. It’s the community making these specific ones. You want to download that so it can do everything and the kitchen sink. Your kitchen sink is leaking. I have an app for that to fix that.
Per: Imagine an app to download to your robot that does physical things that carry my luggage to the car robot or whatever.
Daniel: Exactly so there would be silly little things. It’s going to be an app that makes the robot snap its fingers. That is what it will do. The community will make the things that it knows it needs because people like me out there that work very hard to make their lives easy and thus it'll hopefully help everybody else.
Per: Very interesting, thank you very much for taking the time. I appreciate you doing an interview.
Daniel: Well, thank you very much for having me.
Per: I hope you liked this episode of the podcast version of Robots in Depth. This episode is produced together with Wevolver. Wevolver is a platform and community providing engineers informative content to help them innovate. It is how engineers stay cutting edge. Aptomica is the founding sponsor for Robots in Depth. Aptomica runs anything in modular robotics. Dream, rent, build. Visit Aptomica.com to connect. I am your host Per Sjöbor.
END OF TRANSCRIPT