Neural MMO 2.0, a massively multi-agent environment for RL research

Neural MMO 2.0 is a massively multi-agent environment for reinforcement learning research. It enables populations of agents to be simulated in procedurally generated virtual worlds. Agents must forage for resources to stay alive, mine materials to increase their combat and task completion capabilities, level up their fighting styles and equipment, practice different professions, and engage in trade based on market demand.

The environment also supports a flexible task system, allowing many potential goal and reward states.

This talk will describe the environment, and the engineering decisions that make it performant.