Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning to Play Foosball: System and Baselines

Janosch Moos, Cedric Derstroff, Niklas Schröder, Debora Clever

TL;DR

This work stages Foosball as a versatile platform for advancing scientific research, particularly in the realm of robot learning, and spotlightes the automated Foosball table as an invaluable tool, possessing numerous desirable attributes, to serve as a demanding learning environment for advancing robotics and automation research.

Abstract

This work stages Foosball as a versatile platform for advancing scientific research, particularly in the realm of robot learning. We present an automated Foosball table along with its corresponding simulated counterpart, showcasing a diverse range of challenges through example tasks within the Foosball environment. Initial findings are shared using a simple baseline approach. Foosball constitutes a versatile learning environment with the potential to yield cutting-edge research in various fields of artificial intelligence and machine learning, notably robust learning, while also extending its applicability to industrial robotics and automation setups. To transform our physical Foosball table into a research-friendly system, we augmented it with a 2 degrees of freedom kinematic chain to control the goalkeeper rod as an initial setup with the intention to be extended to the full game as soon as possible. Our experiments reveal that a realistic simulation is essential for mastering complex robotic tasks, yet translating these accomplishments to the real system remains challenging, often accompanied by a performance decline. This emphasizes the critical importance of research in this direction. In this concern, we spotlight the automated Foosball table as an invaluable tool, possessing numerous desirable attributes, to serve as a demanding learning environment for advancing robotics and automation research.

Learning to Play Foosball: System and Baselines

TL;DR

This work stages Foosball as a versatile platform for advancing scientific research, particularly in the realm of robot learning, and spotlightes the automated Foosball table as an invaluable tool, possessing numerous desirable attributes, to serve as a demanding learning environment for advancing robotics and automation research.

Abstract

This work stages Foosball as a versatile platform for advancing scientific research, particularly in the realm of robot learning. We present an automated Foosball table along with its corresponding simulated counterpart, showcasing a diverse range of challenges through example tasks within the Foosball environment. Initial findings are shared using a simple baseline approach. Foosball constitutes a versatile learning environment with the potential to yield cutting-edge research in various fields of artificial intelligence and machine learning, notably robust learning, while also extending its applicability to industrial robotics and automation setups. To transform our physical Foosball table into a research-friendly system, we augmented it with a 2 degrees of freedom kinematic chain to control the goalkeeper rod as an initial setup with the intention to be extended to the full game as soon as possible. Our experiments reveal that a realistic simulation is essential for mastering complex robotic tasks, yet translating these accomplishments to the real system remains challenging, often accompanied by a performance decline. This emphasizes the critical importance of research in this direction. In this concern, we spotlight the automated Foosball table as an invaluable tool, possessing numerous desirable attributes, to serve as a demanding learning environment for advancing robotics and automation research.
Paper Structure (19 sections, 5 figures)

This paper contains 19 sections, 5 figures.

Table of Contents

  1. INTRODUCTION
  2. BASICS
  3. Adversarial RL
  4. Related Work
  5. Learning to Play Foosball
  6. Our Autonomous Foosball System
  7. Training Setup
  8. The Training Environment
  9. The Observation Space
  10. The Rewards
  11. EXPERIMENTS AND RESULTS
  12. Base Skills
  13. Blocking
  14. Scoring
  15. Self-play
  16. ...and 4 more sections

Figures (5)

  • Figure 1: Our autonomous Foosball system in a human versus machine setting for robust and adversarial robot learning research.
  • Figure 2: Shown is the CAD model of the actuation for the autonomous Foosball system. A BLDC motor operates a belt drive system with a carriage creating a prismatic joint. A second BLDC motor is mounted onto the carriage and connected to the rod to add a revolute joint.
  • Figure 3: Shown is an example of synthetically generated data used to train the detection algorithm YOLO used in this work. The bounding boxes are created from a set of points projected into the image plane. These points are taken from the underlying CAD model of the figurines.
  • Figure 4: Shown is a subset of the state estimation results from YOLO detection followed by Kalman filtering in comparison the the true values from simulation. All estimates are in pixels or pixels per frame at 60.0 FPS. The white keeper rod is chosen in (a) as representative for position estimation along its prismatic joint. Other rods display similar performance. In the center (b), the x- and y-position estimate of the ball is plotted. Lastly in (c), the ball velocity in x-direction is shown. We have omitted the y-velocity for clarity as it yielded similar results.
  • Figure 5: Shown are the learning curves on our example tasks. The top row consists of the results on the base skills, while the bottom row illustrates the results of the self-play training. In each case, the left subplot is in terms of success rate, and the right subplots shows the reward over learning episodes.