[Paper Review] Predicting opponent team activity in a RoboCup environment

Selene Báez Santamaría|arXiv (Cornell University)|Mar 4, 2015

Artificial Intelligence in Games6 references3 citations

TL;DR

This paper proposes a low-complexity Markov model to predict opponent team formations in the RoboCup SSL environment by modeling player states via position, velocity, and team centroid, with transitions learned from vision data and stored efficiently in a hash table. The method achieves real-time prediction with minimal computation by using transition probabilities and confidence scores derived from individual player probabilities, significantly reducing time and space overhead.

ABSTRACT

The goal of this project is to predict the opponent's configuration in a RoboCup SSL environment. For simplicity, a Markov model assumption is made such that the predicted formation of the opponent team only depends on its current formation. The field is divided into a grid and a robot state per player is created with information about its position and its velocity. To gather a more general sense of what the opposing team is doing, the state also incorporates the team's average position (centroid). All possible state transitions are stored in a hash table that requires minimum storage space. The table is populated with transition probabilities that are learned by reading vision packages and counting the state transitions regardless of the specific robot player. Therefore, the computation during the game is reduced to interpreting a given vision package to assign each player to a state, and looking for the most likely state it will transition to. The confidence of the predicted team's formation is the product of each individual player's probability. The project is noteworthy in that it minimizes the time and space complexity requirements for opponent's moves prediction.

Motivation & Objective

• To predict the future configuration of an opponent team in the RoboCup SSL robotic soccer environment.
• To minimize computational and storage overhead in real-time opponent behavior prediction.
• To model team-level dynamics using a Markov assumption based on current team state and player positions.
• To enable fast inference during gameplay by precomputing and storing transition probabilities.
• To quantify prediction confidence using the product of individual player transition probabilities.

Proposed method

• The field is partitioned into a grid to discretize player positions and form states.
• Each player's state is defined by its position, velocity, and the team's centroid (average position).
• A hash table stores all possible state transitions with their observed frequencies from vision data.
• Transition probabilities are learned by counting state transitions across all players, regardless of identity.
• During inference, the current vision package is mapped to states, and the most likely next state is selected using precomputed probabilities.
• Prediction confidence is computed as the product of individual player transition probabilities.

Experimental results

Research questions

• RQ1 How can opponent team formations be predicted efficiently in real time within the RoboCup SSL environment?
• RQ2 What is the minimal state representation that captures team-level dynamics without tracking individual players?
• RQ3 Can a Markov model with low storage and computational cost effectively predict team behavior?
• RQ4 How can prediction confidence be quantified using probabilistic transitions?
• RQ5 What is the impact of using team centroid and velocity on prediction accuracy?

Key findings

• The method achieves real-time prediction with minimal time and space complexity due to efficient hash-based storage of state transitions.
• The prediction confidence is computed as the product of individual player transition probabilities, providing a measurable reliability score.
• The model learns from raw vision data without requiring player identity tracking, reducing complexity.
• The use of a team centroid improves the generalization of team-level behavior modeling.
• The approach enables fast inference by offloading computation to a pre-processing phase during training.
• The Markov assumption simplifies the problem by making predictions depend only on the current state, reducing model complexity.