GO2008: First success

Yesterday and today were the first competition tests. We call a run for a specific competition task a test (or alternatively challenge) in our league.

Yesterday we had the introduce test. In that test the robot shows it's features and capabilities as a quick introduction. This is basically for the audience and the result is voted for by the other teams. The next challenge was Fast Follow. In this challenge the robot has to follow a human. This year the difficulty was raised by having two teams in the arena at the same time competing for speed and follow accuracy. The robot has to track the human and not accidentally follow the other team and a team has to try to finish the parcours faster than the others. In both tests we achieved the highest possible ratings with 891 and 1000 of 1000 points respectively. The Fast Follow was improved this year by my team member Masrur Doostdar. Up to now we only used our laser to recognize the legs of humans and to track these to follow. This year we use a vision application I wrote to detect the team logo on our team shirts and use it as an additional feature to improve the tracking, as shown on the picture to the left.

Today we had the Lost'n'Found test and the Open Challenge. In the Lost'n'Found 3 objects out of a set of 10 is put somewhere in the environment on the ground or a table and the robot has to explore the arena and find it. This is a really tough test since the object recognition is not trivial. The lightings conditions vary in the arena, the objects are non-uniform and you see them from different angles and distances. This is the reason why this went quite bad. We were the only team to find anything, but only one object. Another important "feature" that we had was to leave the arena before the timeout to get extra points for demonstrating that the robot is able by itself to drive through the pretty narrow door to the outside. This tests only gave us 150 points, but we were the only team to score at all.

After this there was the Open Challenge. In that tests every team is allowed to come up with an own scenario and demonstrate the specialties of what they can do. You can see this as a free style test. We basically demonstrated a more speedy version of our final challenge of last year. In that competition you teach the robot your cup (one out of three). It remembers your cup and then goes to a waiting position. If you then tell the robot that you forgot your cup, it's going to offer you the service to get the cup. Of course lazy programmers take this (think of the cup as a cup of coffee). The robot then starts looking at well-known positions for the cup. It uses a stereo camera for recongizing and localizing the cup in 3D space and grasps it with a gripper (see the photo to the right). It then brings it to the asking person (up to now at a pre-defined location) and puts the cup on the table. The guys from the University of Koblenz showed their software from the rescue league to detect humans. Unfortunately it detected the same human ever and ever again. The folks from IAIS showed their face recognition and recognized people. Here the criterion of a mobile robot was not quite met, but otherwise it worked. An unknown stranger was recognized as being unknown. The robot said that it remembered that face, but it was not demonstrated later on - the person was not put in front of the robot a second time. The Metor-M team showed their FPGA-based pendulum platform. This was basically what was shown in the introduce task. The FPGA board by itself is pretty cool by itself. In this test we got 1638 points out of 2000, Koblenz got

After these tests we are currently first with 3679 points, followed by IAIS with 2863, Homer from Koblenz with 2279 and Meteor-M with 1330 points. This may change soon since we decided to skip the next stage 1 tests for now to be able to work on the tests for the stage 2. We tend to be most productive at the event, and not very much before the event...

GO2008: Hack your robot

We finished the setup of the robots yesterday (I can't remember when that was, a sense of time is always the first thing that is lost during a RoboCup competition). The maps are built, object have been trained, navigation is working.

The AllemaniACs are particularly proud of their robust Monte Carlo-based localization approach using the data from the integrated laser range finder and the good navigation and especially collision avoidance. You literally can jump in front of the robot when it's moving and it will smoothly brake and drive around you. Having these robust methods allows us to concentrate on the RoboCup@Home-specific tasks instead of having to care about the low-level system.

Currently we are implementing and fine tuning our applications for the different tests. Tests are for example a simple introduction, where the robot introduces itself and its features; another challenge is "Fast Follow". In that test there are two robots and two humans of different teams in the arena. Both humans will go from one side of the arena to the other in opposite direction and the robot has to follow his human team member. The robot has to keep track of the human even if the path crosses the other robot and human. Another test is lost and found where three objects out of ten are placed in the arena and the robot has to find and identify them. For all the tests see the RoboCup@Home Rulebook 2008.

To the left in the image you see our mid-size league soccer robot. In this league teams of usually 5 or 6 robots each play soccer against each other on a field of about 12m x 18m with a standard human soccer winter ball. On the screen you can easily spot the stereo camera on the robot used for close range ball and obstacle detection (though not yet used). Just above that the dark red device is another camera pointing to a mirror located above the camera producing images of the size of 1000x1000 pixels. By looking from the bottom towards the mirror you get a roundview around the robot and thus in every frame a perception of the full 360° environment. The tanks on the robot are for pressurized air, they power our pneumatic kicking device, which can shoot to the left, right and front on the ground and in a parabola. During the game the cables are removed of course, but this robot was offloading images at that time that had been taken for offline testing later on. It has three so-called omni wheels that allow the robot to drive in any direction at any time without turning.

In the background you see the new standard goals of this league without any color markers and with a net (last year they had an opaque yellow or blue background).

GO2008: Setup

Today we - the AllemaniACs RoboCup team arrived in Hannover for the German Open 2008 RoboCup competition (GO2008).

To Fedora Planet I should probably shortly introduce myself. I'm a computer science student working for a couple of years in a robot lab now developing mobile robots - particularly robots that play robot soccer fully autonomous (no remote control!) and do service robotics in a home environment. I responded to Paul W. Frields post about Fedora stories that we are using Fedora to power our robots. This caused interest for example in FWN 128 and so I'm here now to tell you more about robots, RoboCup and Fedora in that context.

This time we are going to participate in the RoboCup@Home competition to defend our title and will do intensive testing with our new mid-size robot platform platform.

Currently we are setting up and preparing the robots. This involves creating a map of the environment for the robot to localize itself, taking pictures and calibrating object recognition code. It also involves a lot of coding and hacking to tweak and fix the robot's behavior and perception. Right now this means that we are integrating code for face recognition that has be developed by Vaishak over the last couple of months...

We are part of the Hannover Fair. The RoboCup is located in hall 25, so if you happen to be in Hannover come by and have some fun, seeing robots autonomously playing robot soccer, servicing in a home environment or rescuing (fake) people in a disaster area.

On the image to the left you see part of this year's RoboCup@Home arena with our robot inside. As you can see the robot has three camers (one is currently not mounted on the top), a camera with a pan/tilt unit on the lower level and a stereo camera on top. It has two wheel chair motors powering with 6 horse powers, which makes the robot pretty speedy, even with it's almost 70kg of weight. Additionally it has a 360° laser range finder for collision avoidance and localization. And of course there is the arm, our actuator to manipulate the world holding one of the most important ingredients of robotics: duck tape.

Did you port already?

After the switch to FreeBSD of kernel.org I'm glad that I have just ported our robot software to FreeBSD ;-)

Gems in the Linux standard headers

I was just looking around the headers for adding inotify support to a plugin for our robot software when I found this in /usr/include/linux/magic.h:

#define FUTEXFS_SUPER_MAGIC     0xBAD1DEA
#define INOTIFYFS_SUPER_MAGIC   0x2BAD1DEA