E66: Louise Poubel, Software Engineer at Open Source Robotics Foundation – Interview

October 18, 2016


This interview with Louise Poubel is all about software for robotics. Louise is a software engineer at the Open Source Robotics Foundation (OSRF) located in the San Francisco Bay area. OSRF is helping make robotics more accessible and easier to build than ever before.

Louise has a fascinating background. She grew up in Brazil, studied in Japan and did her masters in Poland and France. She’s currently working on different projects at OSRF including a project called Gazebo, which is a robot simulation platform.

Louise shares an exciting and hopeful future about the development of robotics.

Some other things we talk about:

-Why did you study in Japan?
-How did you get involved with robotics?
-What is Gazebo? How is it different from what else is out there?
-Louise describes a case study where one robotics firm saved many years using OSRF software platform. It was shocking how much time they saved.
-What type of robot is Louise excited for?


David Kruse: Hey everyone. Welcome to another episode of Flyover Labs and today we are lucky enough to have Louise Poubel with us. And Louise is a Software Engineer at the Open Source Robotics Foundation and she has quite a fascinating background too. She is currently working on different projects at OSR including a project called Gazebo, which is a robot simulation platform. So I invited Louise on the show to hear more about her background and also hear what she is up to at OSR and what they are dealing with by providing this powerful robotic tools to anywhere in the world, and this could really change how robotics are developed, tested and deployed throughout the world. So Louise, thanks for coming on the show today.

Louise Poubel: Thank you for having me.

David Kruse: Definitely. And so yeah, you have a really interesting background. Can you tell us a little bit about it?

Louise Poubel: Well, I’m Brazilian and I grew up in Brazil and I went to college in Japan, so I think that’s one of the things that people find interesting.

David Kruse: It was in Japanese too.

Louise Poubel: Yeah, it was all in Japanese. I was studying with Japanese students like doing everything like the way they do, all the courses in Japanese, yeah.

David Kruse: Did you know Japanese before going or…

Louise Poubel: No, it was part of the program I went. I went with a scholarship from the Japanese government and they just pay you to stay five years in Japan, one year studying Japanese and after that one year you join college together with the Japanese students.

David Kruse: Oh my goodness, that’s a child on fire. That’s not a lot of time how you learn Japanese, but…

Louise Poubel: Yeah, it’s quite tight. But it was enough for me to graduate I guess. So yes, I studied Electro Mechanical Engineering there at Chiba University.

David Kruse: Okay, got you, got you.

Louise Poubel: And then after that I went to Europe also with a scholarship, this time from the European Union and where I did a Master in Robotics. So I studied one year in Poland, one year in France and that’s where I started with robotics.

David Kruse: Got you, and so where did you learn English.

Louise Poubel: Well in Brazil we learn English a little bit at school. So yeah, I did like some English schools in Brazil, but I really learnt when I was in Japan, because my English was better than my Japanese and that was the first time that I had to use English and that’s where it really took off.

David Kruse: Oh! Interesting, all right. So do you know those three languages then, Japanese, English and…

Louise Poubel: My Japanese is not as good as my English and I’m forgetting. Ever since I moved out of Japan I haven’t had the chance to practice this much. So I’m losing my Japanese very quickly.

David Kruse: I bet. Well, I had friends who had learnt it and it’s a lifelong endeavor they say to become really fluent. But all right, so after you got your masters, what did you do?

Louise Poubel: So yeah, I got my masters and then right after I finished I started an internship, a remote internship with the company that I work at right now, the Open Source Robotics Foundation. It was through a program for – called Outreach Program for woman. So it’s for – to get women involved in Open Source and it was remote. I didn’t have to come to the U.S. to work with the Open Source Robotics Foundation. It’s all online, open, and it was a great experience. For three months I was working remotely and making contributions to the project and after that I just kept working on them, kept working on them until eventually I was hired to come here to California, yeah.

David Kruse: Okay, great and so can you tell us a little bit about OSR?

Louise Poubel: OSRF, we have too many projects here. One is the robots operating systems and the other one is the Gazebo simulator that you mentioned. So they both play well together. So the robots are pretty constant. It’s not really an operating system; it’s more like a framework built on top of other operating systems for people to more easily connect all the bits and pieces of our robots together. So you make all your drivers for your cameras for your motors, everything to kind of speak the same language and use the same tools and that makes the whole development much faster, so that’s the ROS, the Robots Operating System. And the Gazebo is the simulator that plays very well with ROS and it allows you to have your robots in a virtual world and simulates many aspects of the robot, from the sensors, like what kind of camera image would it produce, what kind of IMU data you would get from the collisions that you would have with other robots and other things in the environment. So you can simulate all this without having to strain your physical robot or even if you don’t have a physical robot, you can make a robot that doesn’t exist in a virtual world. So these are the two projects here, yeah.

David Kruse: And so with Gazebo, how do you simulate like a new environment? I mean are there different environments kind of programmed in there or let’s say you wanted to have one here and maybe I don’t know if you can do this or not but like kind of more simulating a warehouse or is it more general than that, that you understand.

Louise Poubel: Like you can – its general enough that you can simulate any kind of environment that you want. It depends on how much effort you want to put into it, it depends on what other people have done before and shared. So like warehouse, if you own a warehouse you can see if someone had shared a 3D model just a usual like CAD model, usually you can import any kind of model into the simulation and you can start from that. You can use existing robots that had been shared. And you can really simulate any robot from a humanoid to a quad copter to an industrial arm. So it’s general enough that it covers all these cases.

David Kruse: Interesting. And so what do you do specifically at OSR?

Louise Poubel: I work mostly on Gazebo rather than ROS and I work mostly on – well, we all – like we are a small team, so everybody does a little bit of everything, but I do mostly user interface, so the graphical user interface where things are going to go, how are people going to interact with the simulation. We are continuously adding more tools for people to be able to do more and more different things with the simulator.

David Kruse: Got you, okay. And before we get involved with OSR, how did you get interested in robotics and engineering?

Louise Poubel: Well, I always kind of liked the way things move. Like I am fascinated by geometric edges, really like you know seeing how things are going to move, where they start with one they are going to stop and for me robotics is basically just geometry applied. You know you have all these links connected by joints and they move in a certain way based on their geometry. So this for me is fascinating. It’s just like having this real world or in if you are using Gazebo just this virtual things that move. Movement is ridicule, so that’s what drew me into robotics.

David Kruse: That’s a really interesting answer, I like it, about movement, nice. Okay, and so within Gazebo can you – so you can simulate a world like. So in theory would it make training a robot much faster? Could you just, do you – I guess if you need the processing power, but thousands of simulations, whereas using these virtual worlds instead of using one or you know five physical robots you could simulate many robots in Gazebo. Is something like that possible?

Louise Poubel: Yes, definitely, yes. There is a lot of people using simulation to create. As you said some situations are hard for you to simulate in the real environment. So if you want your robots to – like if you have a back dated robot and you want it to climb several types of stairs, you don’t want to build all those physical stairs for your robot to try in the real world. So you can just quickly put them together in simulation and try our algorithm and see if it works for a wide verity of stats much quicker than you would do in real life. The same goes with like conditions for drones, if it’s going to be windy, if it’s going to have obstacles in a certain configuration, you can have much more flexibility in the virtual environment.

David Kruse: Interesting, yeah that makes sense and probably cheaper too, at least in the…

Louise Poubel: Yes definitely. Depending on your robot, it can be really expensive to just to get it running, every single time you want to try something new.

David Kruse: And how do you know that the robot in the simulation and the environment is identical to the robot you actually want to build?

Louise Poubel: That’s actually one of the tricky parts, right. You have to use simulation consciously. You have to know that there is a limit to how much simulation is going to be faithful to reality. It’s a good rule of thumb that if your robot works in simulation, well if it doesn’t work in simulation then it’s not going to work in real life. This is something that probably you can believe, but if it works in simulation it doesn’t translate well to reality every single time. Just you can rule out a lot of the crazy cases, but the fine tuned cases it’s really hard to get that, you know the robots was modeled exactly very, very detailed perfectly in the simulation.

David Kruse: So if you had somebody who was trying to build a robot to go up steps like you mentioned and how – I know this is kind of a tough question, but how – stated the laws of simulation within Gazebo, how could they – yeah, could they get the robots to actually start training, so when they actually put it in the physical world the robot would be pretty smart by the time they get to the physical world?

Louise Poubel: Yeah, it depends on – there is always corner cases that you are not able to predict or stimulate that, and like little you know, little problems that might happen in the real life that are harder to simulate, but… For example, for my Masters research I was doing human motion imitation in real time. So a human will be moving in front of a connect sensor, the one that comes – the Microsoft Connect Sensor for the Xbox videogame, yeah. So we were capturing human motion in real time with that sensor and the robots, a tiny little humanoid robot was copying that motion in real time. And I used a lot of simulation for that and every time I could see that if the robot would be balanced doing one foot in simulation and wouldn’t fall on the ground, it would most probably not fall in the real life, so – and I didn’t want, yeah and its troublesome like to be, you need to have someone else there to be protecting the robot in case it falls in real life. So I did a lot of my practicing in simulation before I brought the algorithm to the real robot.

David Kruse: Interesting. It must be exciting the first time you tried with the real robot and actually stands or gets close to standing on one foot.

Louise Poubel: Yeah.

David Kruse: So before OSR and what type of tools were people using at universities to do something similar, and what’s the advantage of using the tools at OSR?

Louise Poubel: So like the – I think before ROS specifically, I think it was much more complicated for teams to get just the base to start working on their robots. So just doing the whole framework and connecting everything. To start doing, so let’s say you have a team who wants to do some new navigation for your robot. They don’t care about the drivers for the motors, they don’t care about the specifics of that camera they are using, you just want something that works, so you can focus on your little detail. And you as a lab as an instruction would waste a lot of time just putting those pieces together until you finally got to the point where you could work on what you wanted to focus on. So ROS kind of gave you that head start. You can forge from what has been done by other people, what has been shared with the community, to quickly put all the pieces together and then you start focusing on the specific things that you want to focus on. So this is something that really I believe has spread up development, especially in universities where you don’t have big things or sometimes you don’t have money to buy expensive software and it can arrange the leverage from these free open source tools. And the same goes for Gazebo. You can have many students working on the same robot at the same time even though the university only has one robot. They all have their own virtual copies of their robot. So I think it was very good for academia and increasingly it’s been good for industry as well. Like the CEO of Singer Robotics, they make a robot that goes into warehouses checking their inventory. He said that they would have spent 22 years developing that robot if they didn’t have ROS and instead they spent 18 months from nothing to robot. So it’s really powerful, yeah.

David Kruse: Yeah and the beauty of it, it only gets more and more powerful as more people use it and contribute back and yeah.

Louise Poubel: Yeah, that’s true. One of the biggest assets is the community is the fact that there is a lot of people using it, for sure.

David Kruse: Interesting. And do you why they – you probably went at the very beginning, but why they decided to go the non-profit route. I mean I love the mission and everything, but you know, it’s also something that could be pretty interesting for a profit company too.

Louise Poubel: Well, we do open source and people can use our products for free. So it’s kind of hard to make a profit out of this. Of course there are models where you can do this, but yeah, I think that the immediate – I was not one of the Founders, I wasn’t here from the beginning, so I’m not sure on the decisions that were made. But yeah, I’m sure it’s the model that makes sense and has been making sense for a while.

David Kruse: Yeah, that makes sense, okay. And what – we’ll see, I’ll start with the – do you have, actually I guess you already gave a good use case. That one was probably the best of 22 years down 18 months. What about the – so that’s for ROS. What about with Gazebo, do you have any good use cases for that or yeah, a story where somebody used it and saved time or were very happy?

Louise Poubel: Very happy. I think the biggest use case for Gazebo so far has been the DARPA Robotics Challenge. I don’t know if you heard about that. It was a big robotics challenge where humanoid robots had to compete in rescue tasks. So they had to go in a course that for a human you would have taken like two minutes. It’s just like driving a vehicle in a short time and then opening a door, turning a valve or walking over some rough terrain and for a human that would be very easy you know if the human could get to that place. But of course the advantage of using robots like with radiation, like at the time they were thinking about the Uchiyama Portland in Japan that there were no robots to send in there. So that was a motivation for the challenge.
And the first phase of the challenge was a virtual competition inside Gazebo. Yeah, yeah, so like things from all around the world, anyone could enter the competition and they would – everybody would be using this very same robot, in the very same conditions and simulation and they would all – so the only thing they were competing with who had the best algorithms, right. It’s not about who is building the best robot, about who can use the robot best. So the competing was all running Gazebo. It was running the cloud so the teams could control their robots from far away and the teams who won the competition, the virtual competition they received the real robot, a physical robot to compete in the physical competition later on and they received funding and all. So the first phase was in Gazebo and out of the 23 teams that were in the finals, 14 teams had used Gazebo actively for their development. So including like many of the best best teams.

David Kruse: I’m curious, do you remember how fast the humanoid finished the track or the…

Louise Poubel: Oh! It was almost an hour.

David Kruse: An hour.

Louise Poubel: Yeah.

David Kruse: Us humans are so smart. No, it’s pretty funny, but I mean look at the DARPA Self Driving Car Competition, just like 10 years ago or 12 years ago whenever that was and how bad they were in now. So you know maybe in 10 years we’ll have humanoids walking all over the place.

Louise Poubel: Exactly, definitely the competition really made the robots be better. Like between the midterm and the finals of the competition there was a big improvement in the robots. So hopefully they will go on it and continue improving from now on.

David Kruse: So how could they get that down to two minutes? I mean just in a single simulation would more, if they had all of Google’s processing power could they get that down do you think or is it more keep tweaking the algorithms, is that more of the issue or is it more training?

Louise Poubel: I think there is probably a lot of improvement to be done on all sides; from the mechanic to the decision making. In the competition it wasn’t fully autonomous. Like there were people who could control the robots, but there was some latencies in the communication. So if you relied purely on the humans giving instructions and no autonomy for the robot, you would waste a lot of time in this back and forth in the communicating. So yeah, in the case of this competition, building more, putting more autonomy on the robot would have gained them some time and of course there is a lot of other things like battery life and things like this that would really help.

David Kruse: Yes, yes, got you.

Louise Poubel: And there is plenty of room today for improvement.

David Kruse: Makes sense, okay. And what’s kind of the five year shorter or longer vision for Gazebo? What else…

Louise Poubel: Gazebo?

David Kruse: Yeah.

Louise Poubel: Well, one of the mid to long term goals that we have right now as a team is to improve the usability of Gazebo. So we really want to make it more accessible for people of different backgrounds. Right now Gazebo is a great tool if you are a computer scientist and you know your way around Linux and you can, you can find your way and for this kind of user it’s great. But we are looking into other users who could profit from Gazebo. For example Mechanical Engineers like myself who might not be so Linux savvy, who might not C++ so well. How can we make Gazebo easier for these people to also you know take advantage of it? So we are really trying to build some tools to make it, like make it easier for you to build robots right now. It’s a lot of text based, you have to write a lot of text to build your 3D model. We are currently working on the robot editors that is point and click, more like a CAD software, yeah and like having a online database of models, so people can share their existing models and more easily put together their world, like we were seeing before and also like some other technical things like making Gazebo run on Windows. You know a lot of non-computer scientists, you know most people use windows and it would be nice to just have Gazebo ready to work on Windows properly.

David Kruse: Interesting, lots of exciting features, that would be great if you like opened it up to lots of different folks. Interesting, okay. And well you kind of answered, I was curious what the text doc was; what language that you were using where you kind of answered with Linux. But I was curious, if somebody wanted to develop let’s say a new environment within Gazebo to simulate a robot in, let’s say like a Cherry Orchard, just throwing it out there, how would they go about doing that? Would it be all text – I guess would it be all text based or how do they make that happen?

Louise Poubel: Right now we have robot editor where you can do most things. You cannot do – you cannot edit all the little details of the robot yet or very easily, but that’s something that we are working actively on. But yeah, it would be mostly writing packs which is like kind of like XML or HTML, that’s the way that you describe their roles, but if you actually want to make things move and send comments to things in your world then you need to know some C++.

David Kruse: Got you, okay so it’s not too bad.

Louise Poubel: Yeah and of course like knowing Linux.

David Kruse: Yeah, wouldn’t hurt. It’s always essential.

Louise Poubel: Yeah, it does help.

David Kruse: All right, so yeah actually building it out isn’t too bad. It sounds like it’s a kind of a mark-up language. But then, for them to actually make it useful, it probably needs a little more technical knowledge.

Louise Poubel: And C++ yeah

David Kruse: Yeah, interesting okay. Yeah and we are almost more done here at the interview unfortunately, but got one or two more questions. I was curious, this is more of a fun question, but if you could build one type of robot right now to do something, what would it be? If you think about this, I don’t know if you think about it, but I do.

Louise Poubel: I think about it all the time yeah. And as I was saying, if I had an answer right now I would go home and just start building that robot right now, yeah.

David Kruse: No, it’s true.

Louise Poubel: I’m sorry.

David Kruse: Oh! No its true like, right it’s hard to know exactly what would be the most usually as far as the robot, yeah. But yeah, so you don’t necessarily have an answer or you are not sure?

Louise Poubel: Yeah, well I have a vague idea that I would like. I really think the robots can do a lot of repetitive tasks that you know humans don’t need to be doing. So things around the house or in their daily lives that really we could just delegate to our robots and so I would to do a robot that would help people in their daily lives and free their time like this, like doing their laundry and washing the dishes and you know grocery shopping and stuff like that, self driving cars. And also I’m really interested in empowering people. So if I you know could work on making a robot that would facilitate manufacturing, I really like 3D printing and I really think it has the potential to empower power. So you know bringing factories to little community and making them as dependent on other big corporations, that’s would be cool. So I had only vague ideas. I don’t have the specific robot like.

David Kruse: No, I like the idea. Well, especially with the home robots, it’s something that could you know completely change. Like the iPhone when it came out was cool, and it kind of helped people definitely in certain growths, but man it doesn’t necessarily save you time. And so if you had like a laundry robot or a cooking robot people would stay in their homes longer and you could save a lot of time and oh my goodness that would be nice.

Louise Poubel: That would be great.

David Kruse: So could you simulate a laundry environment in Gazebo?

Louise Poubel: What kind of environment, I’m sorry?

David Kruse: Or like a robot doing laundry.

Louise Poubel: Doing laundry. I think some – there has – I’ve seen people like folding cloths on like using robots to fold cloths in Gazebo. It’s kind of hard because we don’t simulate fabric very well. So it’s kind of like a shirt that is made out of like plates that’s folded on the joints. It’s a very rigid body Gazebo right now, but yeah people do a lot of like cooking environment. There is like, if you download Gazebo right now you can put on a kitchen there and you can put like plates and things like that.

David Kruse: Oh wow! Okay, interesting. And I know you are creating a virtual environment with the Gazebo. Have you put it in virtual reality, you know like an actual VR headset like Oculus or like something similar?

Louise Poubel: Oh! Yeah.

David Kruse: Have you done that?

Louise Poubel: Yeah, we did some demos. We support right now that the Oculus Rift, but this like it’s been a couple of years that we have been supporting it and when we did demos for conferences and things like this, people usually can like control the robots from the point of view of the robot in the simulation. Yeah, we are also starting to support the open source VR or SVR set. So yeah, you can be inside Gazebo.

David Kruse: Well, that’s cool. Do you think – I was always curious. Do you think that’s a big help seeing the simulation in like in VR versus just on a computer screen. Like does it help you make better decisions and choices about the robot and how to build it?

Louise Poubel: I’m sure it might help. Well, the first person point of view can be also limiting right. It’s not the same as seeing things from very far away in doubt. But like once this guy came asking me if like that he wanted to show his clients his robots in a factory and he wanted to use the VR for this, like to show how the robot – for them to feel like how it would feel if they bought the robot and put it in his factory. So there is a little bit of that, I guess for the VR in yeah.

David Kruse: Yeah, yeah. No, that makes sense. Okay.

Louise Poubel: I’m not so sure as much for developments. Maybe there is, I mean VR has so much potential right.

David Kruse: Yeah, exactly. Interesting, okay. Well, unfortunately I think that just about does it for this interview. But yeah, it was really good to meet you Louise and hear your story and hear more about what your working on.

Louise Poubel: Nice chatting with you, yeah.

David Kruse: Its really interesting and yeah, I mean the tools your building now are definitely going to help the rest of the world build tools for the future. So it must be pretty exciting going to work. Well, besides all the meetings and the stuff, the overall must be pretty exciting, going to work each day.

Louise Poubel: Yeah, it’s pretty fun.

David Kruse: Nice, okay. Well, yeah thanks again and thanks everyone for listening to another episode of Flyover Labs. I appreciate it as always and we’ll see you next time. Bye everyone.

Louise Poubel: Bye.