The main goal of this work is to facilitate machine learning research for multi-robot systems as they occur in RoboCup, an international scientific robot competition. We describe our software (a simulator patch and scripts) and a larger research dataset from games of some of the top teams from 2016 and 2017 in Soccer Simulation League (2D), where teams of 11 agents compete against each other, recorded by this software. We used 10 different teams to play each other, resulting in 45 unique pairings. For each pairing, we ran 25 matches, leading to 1125 matches or more than 180 h of game play. The generated CSV files are 17 GB of data (zipped), or 229 GB (unzipped). The dataset is unique in the sense that it contains local, incomplete and noisy percepts (as sent to each player), in addition to the ground truth logfile that the simulator creates (global, complete, noise-free information of all objects on the field). These data are made available as CSV files, as well as in the original soccer simulator formats.