Avalon: A Benchmark for RL Generalization Using Procedurally Generated Worlds - Data Supplement

This is the data supplement for <em>Avalon: A Benchmark for RL Generalization Using Procedurally Generated Worlds</em>. This version corresponds to the latest round of human player data collection. <strong>avalon__agent_eval_worlds.tar.gz</strong> contains all the evaluation worlds on which agents are scored <strong>avalon__human_eval_worlds.tar.gz</strong> contains all the evaluation worlds on which humans are scored (same as agent worlds, just some minor visual differences e.g. the colors are slightly modified for human comfort) <strong>avalon__human_data.tar.gz </strong>contains all the metadata, actions, inputs, and observations for the human players <strong>avalon__human_scores.json</strong> contains the scores for the human players