University of Hamburg - to research, to teach, to educate and form, to homepage
Hub for Crossdisciplinary Learning
Hub for Crossdisciplinary Learning

Photo: UHH/von Wieding

Hub for Crossdisciplinary Learning

Hamburg Bit-Bots: Learning Robots – Learning by Simulation

The Hamburg Bit-Bots are a student work group in computer science which deals with the development of humanoid robots. The objective is the participation in the annual RoboCup-World Champignon Ship, a highly regarded competition in Robotics.

Will a team of humanoid robots be able to defeat the human soccer world champignon?

This is the question scientists all over the world have asked themselves for more than 20 years by now. In doing so, they conduct important basic research in all areas of robotics and artificial intelligence. We, the student work group Hamburg Bit-Bots, participate in this initiative, the so called RoboCup, too. Every team consists of four autonomous humanoid robots, halftimes last 10 minutes only and everything is a bit smaller. All the same, the games are at least as exciting as those of the professionals.

We mostly construct the robots ourselves and the software is also almost completely written by members of our work group. We work on all topics of robotics – from circuit boards and component development to Computer Vision and inverse cinematic and the generation of movement. The gained insights can be tested at the world champion ship once a year. We even reached the third place this year! However, it is not only about the games of course. Our ideas regularly lead to scientific projects which we present at conferences.

Exchange with other teams

The exchange with other teams across the boundaries of the university is also an important aspect of our research. That is why our complete code is publicly available. We have often cooperated with other teams in the past in order to find solutions for especially difficult problems together.

Improved hard- and software

In this project, we have implemented many improvements in the hardware as well as in the software and were thus able to publish four papers at conferences. In order to spare the hardware of our robots, we have built in elastic parts in the shoulders and torsion springs in the knee, for example. We have improved the control of the motors with a new circuit board which we have developed ourselves. We have redesigned some parts such as the foot plate in order to save weight, for example. This new hardware enabled us to implement a new algorithm which enables our robots to get up faster if they stumble. We were able to reduce the time which needed for this motion from about 8 seconds to under 3 seconds.

Continuous Integration

The second important aspect, we have examined in this project is Continuous Integration. It is about creating an infrastructure which will automatically test the code we have developed and will inform us about mistakes or incompatibilities. Thus, we can ensure that the current version of our code will always work and avoid that a quick, careless change might create big problems. Furthermore, we save time and work force as many mistakes will be detected directly even before we inspect the code.

Computer Vision

The third point which we have worked on is the topic of Computer Vision. For this purpose, we published a data set with over 30.000 annotated pictures and used it to train a new algorithm which is able to recognize relevant objects in soccer such as balls, goals, lines and players. In order to generalize the algorithm to the greatest possible extent, the data set contains pictures with different light conditions, different cameras and different balls. At the same time, we have improved the self-localization of our robots. The algorithm is now based on a particle filter and thus enables us as the first team of the KidSize League (robot’s height max. 80 cm) to play against robots of the AdultSize League (robot’s height ca. 1.80m) and to score a goal even.

In total, we were able to publish four papers this year, as well as reaching the third place in the world championship and the first place in the Brazilian championship. We are very happy about these results and have already started to work on winning the world championship next year.

Student research group

  • Hendrik Brandt
  • Timon Engelke
  • Niklas Fiedler
  • Jasper Güldenstein
  • Jan Gutsche
  • Jonas Hagge
  • Finn-Thorben Sell
  • Sebastian Stelter
  • Florian Vahl

Mentor

  • Marc Bestmann

Two robots in a soccer duel

The research group in an online meeting

The playing field of the BitBots