Team Sweaty

Two configurations were developed for the visual detection of objects on the playing field (ball, goals, lines, opponents). The first variant was provided by the Nimbro team at the University of Bonn. It consists of a Logitech camera with a fisheye lens (approx. 185° field of view), which was mounted on a small motor to rotate the camera head. Separately, a more complex stereoscopic detection system was built using two USB RGB cameras (1280x1024 pixels), which are mounted on a fixed bracket 12 cm apart. The mount can be panned, tilted, and rotated around the central axis of the field of view.

As a starting point, the object recognition component of the Nimbro team’s software for Ubuntu Linux—which runs on the Odroid-XU—was adapted. With this configuration, it provides simple color calibration and subsequently reliably detects the ball, the playing field, and the goals. Another important aspect of this work was becoming familiar with the ROS framework.

This made it possible to establish communication with the Java software, which handles the adaptation and decision-making processes. The ball’s world coordinates could be determined with centimeter-level resolution. Without any optimization, a frame rate of 10 Hz was achieved.

Apart from the Nimbro- and ROS-based analyses on the Odroid-XU, further work was carried out using OpenCV. This was the case, for example, with the geometric calibration of the fisheye lens using the classic checkerboard approach, or with the line detection methods, which rely more on shape than on color features. For fundamental testing, a complete scaled-down playing field with good options for controlling lighting conditions is available. A half-sized playing field was set up at full scale, which is more dependent on ambient lighting conditions.

The torque of conventional drives with a given mass is limited by heat dissipation. Increasing heat transfer automatically helps to increase a drive’s average torque, as long as magnetic saturation has not been reached. This is the case with most conventional drives. One possible drive, for example, is the Dynamixel MX-28R, which is powered by an RE-max-17 motor (manufacturer: Maxon). The thermal resistance between the housing and the air for this motor is 35 K/W; in addition, there is an additional thermal resistance due to the drive’s housing. The thermal resistance between the windings and the motor housing is only 12 K/W. Reducing the thermal resistance to the air by two-thirds would double the allowable heat dissipation. This reduction can be achieved through improved heat transfer to the air. A moist surface enables additional heat dissipation through water evaporation. The drive in question has a stall current of 1.7 A and generates approximately 25 W of heat at maximum current. The motor can be operated at high current only for short periods and only with extended breaks; otherwise, it will overheat and be destroyed. These pauses can be shortened through efficient cooling of the motor [HD1]. The amount of water evaporated in relation to the heat dissipated can be calculated using the mass flow rate, where the heat dissipated is equal to the specific heat of vaporization. If one evaluates this ratio for the drive in question, it results in approximately 1 g of water evaporation per 100 s at the stall current, which, of course, cannot be sustained without pauses.

Sweaty's computing and communication architecture consists of three levels:

A high-performance mini-PC with an Intel i7 processor serves as the main computer. The operating system is Linux. The decision-making programs are written in Java, and the programs for visual environment recognition are written in C++ and Python. 

A main communication controller collects and distributes data between the main computer and the distributed microcontrollers and performs preliminary data analysis. The main communication controller is based on a 32-bit ARM Cortex-M4 and was developed and manufactured in-house.

The distributed microcontrollers (ECUs) are also based on 32-bit ARM Cortex-M4 microcontrollers. They are built into the servos, with additional ones located in the legs and near the head.

Communication between the main computer and the main communication controller is implemented via USB 2.0; communication between the main communication controller and the distributed ECUs takes place via EIA-485 at a clock frequency of 2 MHz. A dedicated communication protocol is used to ensure a standardized exchange of information, describing, among other things, how each ECU reports its current status to the central unit.

Two physical EIA-485 communication networks are required to ensure sufficient data throughput.

In addition to the servos’ ECUs, a total of 4 ECUs are planned: one per foot, one near the center of mass, and one in the head. The behavior of the ECUs is simulated using LTSpice and designed using EAGLE. All ECUs are based on 32-bit ARM Cortex-M4 microcontrollers.

Our current Sweaty uses various servos:

• 12 Maxon BLDC motors with lead screw drives (feet, knees, hips, back)

•  4x Maxon BLDC motors with planetary gearheads (arm rotation, hip rotation)

•  8x Dynamixel servos (arms and head)

•  8x linear servos (fingers)

With the exception of the Dynamixel servos, all servo drives (electronics and software) were developed in-house by the Sweaty team. 

In our previous model, we used only commercially available servos (manufacturers: Dynamixel, Volz).

To achieve sufficient movement speed with high torque, the digital servos were overloaded. The motor windings are cantilevered, so overheating leads to immediate motor failure.

A thermal model of the motors was developed to continuously calculate the temperature of the windings in each individual motor. As soon as a threshold temperature is reached, Sweaty’s movements are reduced to a minimum, or he remains stationary until the motors have cooled down again—Sweaty’s behavior is therefore comparable to that of a real soccer player.

The force sensors are to be mounted under the feet. Of the three inertial systems, one sensor is located near the center of gravity and the other two are in the feet to detect the Earth’s surface. For simplicity, all three inertial systems have the same design. Each of the three inertial systems consists of three accelerometers, three gyroscopes, and three magnetic field sensors from STMicroelectronics. Data acquisition and analysis are performed by our electronic control units (see component architecture).

The software team’s main goal is to reuse as much of the 3D simulation’s program code as possible. The code’s component-based architecture makes this much easier. To give a rough idea of the success, we can say that approximately 50 classes for Sweaty were integrated into the framework. These include, among other things, a specialized inverse kinematics calculation, interface classes for ROS, a specialized component creation process, and the parameterization of certain behaviors. There is also a specialized field meta-model for the various field sizes in the Humanoid League.

The number of Nao-specific classes is also approximately 50, and the number of shared classes is well over 500 (excluding tools; runtime only). Of course, during the development process, some of the 500 classes had to be adapted and modified to make them usable for a humanoid robot.

By using the code components from the 3D simulation, we were also able to reap numerous benefits in terms of the toolset. We were able to adopt magmaDeveloper, the behavior editor, and several other smaller tools needed for the development and testing of such software for a bipedal robot.

Sweaty is simulated in the “Simspark Simulator.” This simulator is used for the 3D soccer simulation league.  It is based on ODE and supports gravity, stiffness, body dynamics, joint connections, maximum joint torque, attitude gyros, force-resistance sensors, accelerations, disturbance models, camera simulations, and much more.

To reduce weight, the robot's load-bearing structure was made of aluminum and carbon fiber composites. Further weight reduction was achieved through a special manufacturing process: for example, the head and feet were produced using selective laser sintering (SLS), a process that allows even complex three-dimensional structures to be manufactured.Creo 2.0 was used as the CAD system, and Creo 2.0 and ANSYS were used for FEM analyses.

The team consists of professors, research assistants, doctoral candidates, and students from the divisions of Maschinenbau, Process Engineering, Electrical Engineering, Medizintechnik, Informatik and Media Studies. It is only through the collaboration of all these departments that Sweaty has been so successful in the RoboCup.

• Prof. Dr.-Ing. Ulrich Hochberg
  Team Leader, Sweaty

• André Friedrich, M.Sc.
  Design and Kinematics

• Fabian Schnekenburger, M.Sc.
  Electronics, Software, and Vision

• Nils Jahn
  Electronics

• Lucas Schickl, B.Eng.
  Design

• Prof. Dr. rer. nat. Michael Wülker
  Vision

• Steffen Schmidt
  Electronics

• Michael Sattler, M.Sc.
  RoboCup Technical Committee / Rules

• Martin Burkart, B.Eng.
  Design / Simulation

• Ludovic Letang
  Mechanical Engineering / Development

• Raphael Koger, M.Sc.
  Electronics / Programming

• Matthias Neudorf
  Design / Development

• Matthias Sebastian Niederhofer, Dipl.-Ing. (FH)
  Organization

• Lars Lehmann
  Game Controller, Communication

• Igor Tropmann, M.Sc.
  Electronics / Programming

• Waldemar Frei
  Design / Cooling

• Yuri D'Antilio

• Armin Dietsche, M.Sc.
  Mechanics / Kinematics / Cooling / Student Support

• Michael Obrecht

• Ulrich Kevin Tankeu Tchakounte
  Design

• Aleksandre Zakaroshvili
  Design

• David Stefan Zimmermann
  Decision

• Sneha Venkataramana
  Vision

• Efstratios Tziallas
  Vision

• Mahdi Sadeghi
  USB Communication, Sweaty

• Christoph Plschek
  Electronics

• Samuel Oesterle
  Head

• Ulrich Messner, B.Eng.
  Design 

• Stefan Großmann
  Artificial Intelligence

• Luisa Andre
  Step Planning, Publicity

• Alexander Derr
  Kicks

• Denise Ehret
  Simulation

• Michael Fehrenbach
  Cooling

• Prof. Sabine Hirtes, M.A. in Design 
  Motion Capture

• Julian Hohenöcker
  Heterogeneous Robot Types

• Sebastian Jung
  Design

• Stephan Kammerer
  Rolling Behavior

• Maximilian Krög
  Training, Goalie, Penalty Shootouts

• Muse Seyoum Teklemarriam
  Vision

• Thomas Münch

• Prof. Dr.-Ing. Axel Sikora
  Programming

"SWEATY's two hands" are the "Brunel Hand 2.0" from Open Bionics, and "SWEATY's two arms" are the “Reachy Arm Module” from Pollen Robotics SAS, used under the Creative Commons Attribution-ShareAlike 4.0 International License.