Dr. Dennis Hong talks about Humanoid Robots
RoMeLa is on a roll. This summer, the Robotics and Mechanisms Laboratory at Virginia Tech, led by Dr. Dennis Hong, took home five major awards from the international robot soccer competition RoboCup 2011. The small humanoid robot DARwIn-OP that they developed won first place in the Kid Size league, and this open source robot is quickly gaining worldwide popularity. Their bigger bipedal robot CHARLI-2 won the Best Humanoid Award (Louis Vuitton Humanoid Cup), making them the first U.S. team to win this title. Moreover, the group is now developing the world's first 2-legged fire fighting robot named SAFFiR for the Navy that will autonomously help humans put out fires on a ship. (The group is also developing a car for blind drivers, but we're going to save that for another story.) We asked Dr. Hong about the past, present and future of his group's humanoid robot research. (Photo credit: Virginia Tech)
Q. First I want to ask about your victory in the RoboCup Humanoid Kid-Size League with the DARwIn-OP. The match between you and the Darmstadt Dribblers was fascinating to watch. What do you think was the key to your victory?
A. Our kid size DARwIn team is a collaborative team between Virginia Tech and University of Pennsylvania. We at Virginia Tech (RoMeLa) are the experts on humanoid robots and system integration, and thus lead the platform development. Prof. Dan Lee at the University of Pennsylvania (GRASP), who is well known in the field of machine learning, and his team lead the software development. We had the best hardware platform with DARwIn-OP, so we decided to work with Dan to bring the best of software and the best of hardware together. It was a beautiful story of successful collaboration.
Photo: Team DARwIn at RoboCup 2011 (Credit: Virginia Tech)
We know DARwIn-OP's system inside out because we are the ones who developed it. When you develop software, you actually have to know a lot about the hardware. That’s the advantage we had over other teams.
Another reason for the success, I think, is that the DARwIn-OP is a brand new platform whereas for example the Darmstadt team has been using their platform for many years. So I think their platform is starting to see its age. So as you can see in the match, the first half and second half it was 5 to 5 and then after that, during the overtime, it looked like their hardware started to get tired. The motors overheat and then things get “shaky.”
Another interesting side note. I do know that a number of teams are very interested in DARwIn-OP and I believe they are planning to use DARwIn-OP as their platform for RoboCup 2012. This year, the Darmstadt team actually brought one DARwIn-OP unit and they used it in some of the earlier games. We could see that they were testing it. So technically we were not the only team that had DARwIn- OP at this year’s RoboCup.
Q. Can we step back to the beginning of this project? How and when did your research project on DARwIn-OP that was funded by NSF start? And how much was the grant?
A. The DARwIn project started in 2004 without any funding. When I first joined Virginia Tech, I had some startup funds so I started a miniature humanoid project. I’ve been interested in many types of locomotion. Leg-wheel hybrids, 6 legs, 3 legs…. So naturally the next step was to investigate 2 legged robots and I’ve been very interested in human locomotion. I thought the best way to study how humans walk is to build a humanoid robot and try to make it walk. And during that process we’ll get a better understanding about the dynamics and controls of human walking. Then in the future we will be able to use this knowledge to develop better prosthetic legs and those kind of things.
So that was my main reason I started the DARwIn project. We were NOT thinking about RoboCup at all at that time. And so DarWiN 0 was developed in 2004, in 2005 DARwIn 1, in 2006 DARwIn 2, and then around 2006 or 2007, Prof. Oskar von Stryk at Darmstadt Dribblers saw our robot and invited us to join RoboCup.
Once we started participating in RoboCup, DARwIn started to become popular. Many research laboratories and universities contacted us saying they wanted to use DARwIn for their research and educational tasks but as a university we couldn’t sell it. So I wrote a NSF research proposal with Prof. George Lee at Purdue University, asking that if you can give us the funds, we’ll develop an open source version of DARwIn and give it out to the universities so that we can contribute to the robotics community. We received a total of 1.2 million dollars for a 4 years project. We are in the second year right now.
DARwIn-Op is a fully open source robot which means that the software AND hardware are open source. All the CAD files, the blueprints and documentations on how to make and assemble it are all online for free, so other teams can build it. I already know that some groups are building the robot themselves, but you can also buy it from a company called ROBOTIS.
Q. The NSF grant is for Virginia Tech and Purdue University. How did you get to cooperate with Purdue?
A. I had been talking with Prof. George Lee about humanoid robots and we thought that we have a fantastic opportunity with our DARwIn series so it was an idea that we came up with together to write the proposal. Purdue’s role in the DARwIn project was to try to figure out what the community needs, to set the specifications and what the universities might be able to use it for different types of research. It can be used for network communications, vision processing, autonomous behaviors, locomotion, mobile manipulation - all the robotics disciplines that people are interested in. Then based on those specifications, we designed the hardware and electronics. We brought in Universtiy of Pennsylvania later on to get some help for the software.
Q. How did the company ROBOTIS get involved?
Q. Does Virginia Tech get any kind of licensing fee from ROBOTIS?
A. I wish (laugh). However, we started this as an NSF research and development project. Whenever ROBOTIS sells anything, I don’t get a dime. There is no licensing fee. Everything is fully open source. In other words, if YOU want to start a company and start building and selling DARwIn-OP, no one is going to stop you. My agenda is, this is my baby, this is my contribution to the robotics community. I truly believe in this robot. This is a great education and research platform. I made it open and want people to use it and benefit from it. At the same time, we want the users to also contribute to the effort and form a community. That’s my main interest.
Q. You are doing what many of the other small humanoid robot companies couldn’t do. They couldn’t build the community. What’s the difference?
A. First of all, I think it’s the passion. I’m a true believer in this. I myself am an educator and a researcher and I know the needs of people in the robotics community. So I’m really passionate about this and am pushing it hard. Of course the NSF grant is helping us because after we built this we were able to give out over 20 units for free to 11 universities in the U.S. to foster a user community. The company, ROBOTIS, is behind this as well. Of course their main agenda is to make money, naturally as a business, and for that reason they are pushing it hard. Additionally our big win at RoboCup of course is not hurting us at all. This was a great opportunity to not only show the world the capability and performance of DARwIn-OP, but also to show that this platform can be used for real research. I consider this as an excellent example of how academia, government and industry can partner together to have a high impact.
Photo: DARwIn-OP (Credit: Virginia Tech)
Q. How do you compare your open source contribution to what Willow Garage is doing with the PR2?
A. Willow Garage is doing a fantastic thing, but I see them more as a software company. PR2 is a hardware that people can use but PR2 is not open source hardware. The software is open source. In our case it’s very different in the sense that both the hardware and software are open source. Personally I am not expecting people to use the software that comes with DARwIn-OP as a product that much. I want people to use the hardware and actually put their own software on it. So my real effort is the open source hardware side, not really on the software side.
Q. So you don’t feel the need to standardize the software in robotics?
A. I think there is a need for that, but I don’t think I am in the position to dictate or contribute and say this is the one software people should use. In other words, there’s actually a group who’s trying to port ROS for DARwIn-OP. If you do that, that’s a standard open source hardware plus software as well.
DARwIn-OP itself uses a standard PC and it can run Windows, LINUX, LabVIEW, C++, etc. You can program it in any language you want. If you can install the software on a PC, you can install it on DARwIn-OP. So I see a lot of people that are ditching the software that comes with DARwIn-OP as a product and use their own. And actually I will be happy to see that happen.
Q. Moving on to CHARLI. How do you compare CHARLI to other humanoids robots like HRP series, HUBO and ASIMO?
A. Of course, everybody agrees that Japan and Korea are way ahead in the full size humanoid research and development in the entire world. United States is jumping in very late. Recently the US is starting to be interested in humanoid robots as well. I think the main reason is, back in 2006, NSF formed a group of expert roboticists in the US and sent them around the world to look at the state of the art in robotics. They published a report called the WTEC report.
That report had a big impact in the general robotics research direction in the U.S. And one of the conclusions of that report was that the U.S. is ahead of many countries in autonomous controls, artificial intelligence and military robotics, but that we were way behind in humanoids. So looking at that, I decided that there was an opportunity here.
So how was CHARLI (Cognitive Humanoid Autonomous Robot with Learning Intelligence) born? I’ve been studying different robots in the world and I noticed that most all of the robots have very similar mechanical configuration. Most, if not all of them, use DC motors, harmonic gear drives, and they are very heavy. We wanted to come up with something different because we didn’t want to copy what everybody else is doing. I wanted to have my own different type of architecture and see if we can come up with a new approach in humanoid design. So my goal was to develop a full size humanoid robot with a goal of achieving very light weight and low cost.
Photo: CHARLI-2 at RoboCup 2011 (Credit: Virginia Tech)
The robot being light weight is important for many reasons. First, this is because of safety reasons. Especially when you are talking about humanoids for domestic tasks - robots living in your home helping with dishes and laundry - safety is of utmost importance. Light weight also translates to lower cost because you don’t need to have powerful actuators.
So how do you make it light weight? CHARLI is actually what we used to call CHARLI-L – “L” standing for light-weight. We came up with some clever mechanical design with gravity balancing springs and 4-bar linkages, and it was very successful. CHARLI -2, which is the second version we used for RoboCup this year, walks very stable. I think this is no less stable than other full size humanoid robots in the world. For example, we are not afraid at all running this autonomously on any stage surfaces as long as it is relatively flat. ASIMO and HUBO can only walk on "certified floors" which need to be really hard, smooth and flat.
We allow CHARLI-2 to walk in public among people without a "spotter" right next to it, since we are very confident about it's walking stability and safety. You do not see this with other similar size humanoid robots in the world. I also often shove it with my hand with quite a bit of force while it walks to demonstrate it's impressive stability. I even broke the metal gears in its waist joint once, as I pushed it too hard. We are that confident. CHARLI -2 walks up to about 0.4 meters per second at this time and I think we can make it even walk faster.
Q. Besides CHARLI, now you have another humanoid project called SAFFiR. Can you explain about this new project? (A prototype of SAFFiR was unveiled for the first time at the recent AUVSI show. Photo credit: John F. Williams/U.S. Navy)
A. Yes. SAFFiR stands for “Shipboard Autonomous Fire Fighting Robot” and it is a multi-million dollar project funded by the U.S. Navy. Basically it is a fire-fighting robot in humanoid form. We used to call the heavier version of CHARLI, CHARLI-H, but that became SAFFiR. CHARLI-L and SAFFiR have very different walking algorithms. The CHARLI-2 we used for RoboCup uses a very similar walking algorithm with most other walking robots in Japan and Korea. It’s the basic ZMP (Zero Moment Point) position control walking. This makes the walking look like "a robot." However SAFFiR has a complete different type of walking approach. We actually put in compliance, or springiness, similar to biological muscles. We developed custom linear actuators that have titanium springs in them. We do force control instead and this is a radically different approach to the conventional ZMP based walking.
We decided to propose a humanoid robot for this particular application for a number of reasons. As the robot needs to navigate the narrow passageways and step over high door sills that you typically see on Navy ships, we felt that bipedal robots would be the mobility configuration of choice. Also since it is a ship at sea that rocks, to stay balanced, we needed legs instead of wheels or treads. Wheels and treads need to continuously rotate with respect to the body, thus it is also very difficult to shield it against fire. With limbs, it is relatively easy – a humanoid robot simply needs to put on a fire-fighting suit developed for humans for example. Of course there are also many associated challenges that come with this, too. SAFFiR will autonomously find and locate the fire, navigate the narrow passage ways on the ship, and work with human firefighters to put out fires. Most of our humanoid robots we have developed were for research and education – now we are developing a humanoid robot that will truly save lives.
Q. What is your goal in humanoids research and where do you think they will be useful in our society?
A. I personally have 3 goals with humanoid robot researches. First, I develop humanoids to learn more about humans. One example is to understand human walking better, as explained before. Second goal is similar to most of the other humanoid researchers from Asian have about humanoids to live in our home doing the dishes and the laundry and clean up the room.
In terms of where the robots will be useful, traditionally we call it the “3 D areas”. Dull, Dirty and Dangerous tasks. I personally call it the “5 Ds”. Dull, Dirty, Dangerous and also Distant and Difficult. “Distant” indicates a situation where the robot is too far away so it will be difficult to remote control it, like the rovers on mars. This emphasizes the recent advancement in autonomous systems like the driverless cars from the DARPA Urban Challenge. “Difficult” emphasizes the recent advancements in artificial intelligence, and thus soon robots will be able to perform tasks that require skills. The SAFFiR robot represents well of these application areas. So my third goal is to develop a humanoid robot that can actually do work for these type of tasks.
Q. What is the biggest challenge to accomplish those goals?
A. There is no single challenge - there are so many. It’s really a system of systems. It’s not just about locomotion, it’s not just about perception, sensing, AI, mobile manipulation….. We really need to do everything. My personal focus right now is the bipedal locomotion part for the base platform. Obviously there are a lot of other research groups working on mobile manipulation and I do not claim to be an expert in that field. Hopefully by the time the lower body of SAFFiR is mature enough we’ll be able to use other people’s technology for the mobile manipulation task. In terms of perception, one of my colleagues at Virginia Tech, Prof. Brian Lattimer and his group, together with Prof. Dan Lee’s group at the University of Pennsylvania are currently figuring out the perception part of SAFFiR because it needs to see things in the smoke.
These days in the field of robotics, you can’t be an expert in everything and thus you need to work with other people and collaborate. I consider myself an expert in mechanical design, system integration and locomotion, and I work with other people to build a system together.
(This interview was originally conducted for a robotics column on the Wall Street Journal Japan.)
Comments