In 2021, TIGERs Mannheim won the RoboCup Small Size League competition with individual success in the virtual tournament, the hardware challenges, and the technical challenge. This paper focuses on our open-source robot hardware which has evolved over the past decade to a highly integrated system. Previous robot generations are outlined and the mechanical and electrical design of generations v2019 and v2020 is explained in-depth.

The new RoboCup Small Size League team TIGERs Mannheim onboard camera software RobotPi requires tools and infrastructure for development and deployment. Therefore in this work a CMake configuration to compile RobotPi on multiple platforms in development and production configuration was created. A custom Raspbian image for deployment of necessary dependencies and configurations to the robots was developed with pi-gen. These steps were integrated into GitLab pipelines for continouus integration. For deployment of RobotPi to multiple robots a usb based automatic debian package installation was implemented. To support development without access to a robot and play field an image simulation tool was created with Blender and a method to load them into RobotPi provided.

In the Small Size League (SSL) soccer competition of the Robot World Cup (RoboCup), the TIGERs Mannheim team uses an image recognition system that recognises the ball directly with cameras on the robots. This reduces the reliance on the league-internal vision system. Based on a previous work, this study report aims at centralising the configuration system and enhancing the detected ball position using undistortion and backprojection. The centralisation avoids deviating configuration states in the multi-robot environment. The former configuration system is modifiedand extended to the robots’ Firmware by introducing a new concept. An undistortion model with few parameters is successfully generated and trained. It combines high performance and accuracy. Additionally, the detected ball position is projected back into the three-dimensional space, which contributes to a more effective control of the robot.The results of this study report reflect an improvement in the overall robustness of the image recognition, although the new system has yet to prove itself in a competitive tournament environment like RoboCup

For a team in the RoboCup’s Small Size League (SSL) it is crucial to accuratelyknow the ball position at any point in time. The regular method of obtaining this information is to receive the location from the leagues vision system. However, besides the limited accuracy of the system, there are situations where it can not detect the ball fully. Therefore, a system to detect the ball is added to the robot in this report. I uses a combination of infrared emitters and receivers to detect the signal that is reflected by the ball. Different strategies for the ball position estimation are shown in this report and a combination of a polynomial fit and a tiny neural network is used to predict the position of the ball up to a few millimeters of accuracy. If however, there are other obstacles in front of the robot, such as an opponent robot,the system looses a lot of its accuracy but is still able to detect the presence of the ball and to estimate its position with an error of about 2 centimeters in each direction.

Implementing the Vision-Blackout Technical Challenge in the Robocup Small Size League.

In the Small Size League (SSL) soccer competition of the Robot World Cup (RoboCup), fast passes between robots are the main tool to overcome the defence of the opponent team and allow for successful goal kicks. Most teams have developed tactics to intercept passes and gain ball control. Therefore, reliable analysis of the current situation and the decision if a pass is feasible are necessary. In this work, different pass rating functions are introduced and tested. Focusing on simplicity and reliability, the pass rater use distances of the opponent robots to the pass line as basis for their ratings. An analysis of simulated soccer games using the new pass raters shows similarly high success rates for passes. Finally, the implemented software allows simple exchange and configuration to adjust to different opponents.

Online adaptation of motion models enables autonomous robots to move more accurate in case of unknown disturbances. This paper pro-
poses a new adaptive compensation feedforward controller capable of online learning a compensation motion model without any prior nowledge to counteract non-modeled disturbance such as slippage or hardware mal-functions. The controller is able to prevent motion errors a priori and is well suited for real hardware due to high adaptation rate. It can be used in conjunction with any motion model as only motion errors are compensated. A simple interface enables quick deployment of other robot systems as demonstrated in Small Size and rescue Robot RoboCup leagues.

TBA

The work described in this report includes the implementation of several new
modules and analysis methods within Sumatra that form a nearly completely
redesigned ooffensivestrategy. One key task described in this report is the handling
of multiple robots in the offensive strategy to support and improve passing and
redirecting to build a dynamic and intelligent offensiveplaying behavior.
The newly created offensive design enables more exible and more scoring dangerous
oensive plays. Furthermore the newly created code is more reliable and easier to
adapt.

Moving robots in the RoboCup Small Size League with fast reaction and low latency is crucial for the overall performance of a team. A good movement control is important to overcome the frictions of the wheels of the omnidirectional robots. Because modeling the friction is quite complex, we applied a learning algorithm based on Relative Entropy Policy Search for learning an optimal movement policy. Experiments in the simulation and with a real robot show that we are able to learn a policy that leads us to a desired target state.