Fuzzy Inference Based Forecasting in Soccer Simulation 2D, the RoboCup 2015 Soccer Simulation 2D League Champion Team

Li Xiao, Xiaoping ChenLecture Notes in Computer Science 2015 · DOI · OpenAlex

Abstract

WrightEagle is the champion of the RoboCup 2D simulation league. This paper introduces the application of a fuzzy inference system (FIS) in WrightEagle, designed to forecast opponent actions without requiring a complete mathematical description of their strategies. By predicting opponent behavior, the agents can make more informed defensive and offensive decisions.

Introduction

In the high-stakes environment of the RoboCup 2D simulation league, 11 autonomous agents must coordinate and compete within a 100ms decision window. Success depends not just on reactive play, but on the ability to forecast opponent intent—specifically, predicting where an opponent might pass the ball.

Building a precise mathematical model of an opponent's strategy is a monumental challenge. The "internal logic" of opposing teams is often a black box, and the simulation environment is too complex for traditional heuristic planning to capture non-linear behaviors effectively. WrightEagle addresses this by employing Fuzzy Inference Systems (FIS) to bridge the gap between raw data and strategic forecasting.

Methodology: Fuzzy Inference in Action

Fuzzy logic allows for "human-like" reasoning by mapping imprecise inputs to actionable outputs. For example, a rule might state: "If the defender is close, the probability of a successful pass is low." This approach bypasses the need for rigid mathematical models of the adversary.

Mamdani vs. Sugeno Models

The research explores two primary FIS methodologies:

  1. Mamdani-Type Inference: A linguistic approach using "if-then" rules with fuzzy sets. While intuitive for design, it requires a computationally intensive defuzzification step.
  2. Sugeno-Type Inference: A more efficient model where the output is a mathematical function of the inputs. This efficiency is critical for real-time soccer simulations.

The Marking Scenario

A core application of FIS in WrightEagle is the "marking" scenario. The system predicts whether an opponent (Opp1) will successfully pass to a teammate (Opp2) while being marked by a defender (Tm1).

The marking scenario involves calculating the time (cycles) for the ball (C1), receiver (C2), and defender (C3) to reach a target point A.

The probability of pass success is determined by calculating the "cycles" (time steps) required for the ball, the receiver, and the defender to reach the target point.

The Inference Process

The FIS processes these inputs through a series of fuzzy rules. By applying logical operators like 'and' (min) and 'or' (max) to membership functions, the system derives a fuzzy conclusion about the likelihood of an event.

The fuzzy inference process uses linguistic rules and logical operators to transform inputs into fuzzy outputs.

Results and Performance

The implementation of FIS led to significant improvements in WrightEagle's performance against top-tier opponents from the RoboCup 2014 competition, including Helios, Yushan, Gliders, and Oxsy.

Evaluation across 2,000 full games demonstrated:

  • Increased Win Rates: The win rate against Helios rose from 58.3% to 66.6%, while the rate against Gliders jumped from 71.4% to 82.2%.
  • Enhanced Defense: The average number of goals conceded per game decreased across all test cases. For instance, against Yushan, the average dropped from 0.77 to 0.64.

These technical advancements were instrumental in WrightEagle securing the 1st place trophy at RoboCup 2015.

Conclusion and Future Directions

While FIS provides a powerful framework for modeling unknown opponents, the manual generation of fuzzy rules remains a bottleneck. Future work aims to automate rule discovery by applying machine learning and big data analytics to historical RoboCup match logs. Additionally, the team plans to expand from single-output models to multi-output systems capable of forecasting entire sequences of play.

Summary updated

Abstract

WrightEagle is the champion of RoboCup 2D simulation league, the latest technic used in WrightEagle will be presented in this paper. This paper introduces the application of fuzzy inference system in WrightEagle. The fuzzy inference system is built without knowledge of the mathematical description of the opponents. With the fuzzy inference system, we are able to forecast the actions of the opponents in some aspects. Once we can forecast the actions of the opponents, we can make corresponding decisions.
BibTeX
@inproceedings{xiao2015fuzzy,
  author = {Li Xiao and Xiaoping Chen},
  title = {{Fuzzy Inference Based Forecasting in Soccer Simulation 2D, the RoboCup 2015 Soccer Simulation 2D League Champion Team}},
  booktitle = {Lecture Notes in Computer Science},
  year = {2015},
  doi = {10.1007/978-3-319-29339-4\_12},
}

Loading related content...

Figure