Supported data for manuscript "Can LLM-Augmented autonomous agents cooperate?, An evaluation of their cooperative capabilities through Melting Pot"

The repository data corresponds to the manuscript titled "Can LLM-Augmented Autonomous Agents Cooperate? An Evaluation of Their Cooperative Capabilities through Melting Pot," submitted to the Artificial Intelligence Journal. This data encompasses a series of experiments conducted with Large Language Model-Augmented Autonomous Agents (LAAs) as implemented in the "Cooperative Agents" repository [add link], utilizing the Commons Harvest substrate from the Melting Pot framework. Each experiment consists of 10 simulations, with summarized metrics, graphs, and CSV data provided for each. In every simulation, each game episode sees the participation of a predetermined number of LLM agents and bots. The LLM agents each perform a high-level action in their turn, continuing until all three agents have completed their actions. Concurrently, the bots move continuously, executing a move for every two moves made by any agent (a high-level action typically involves multiple movements). Each simulation concludes either after reaching a preset maximum number of rounds (usually 100) or prematurely if all apples in the environment are consumed. We extract a series of indicators and metrics from each simulation to assess the individual and cooperative performance of the agents.