SAFE: GridWorld

Simple and Fast Environment for Embodied Dialog Agents


Fast and scalable reinforcement learning environment for the IGLU competition at NeurIPS 2022. The env represents an embodied agent with the ability to navigate, place, and break blocks of six different colors.

IGLU is a research project to bridge the gap between reinforcement learning and natural language understanding in Minecraft as a collaborative environment. It provides the RL environment where the goal of an agent is to build structures within a dedicated zone. The structures are described by natural language in the game’s chat.


  title={IGLU Gridworld: Simple and Fast Environment for Embodied Dialog Agents},

  author={Zholus, Artem and Skrynnik, Alexey and Mohanty, Shrestha and Volovikova, Zoya and Kiseleva, Julia and Szlam, Artur and Cot{\'e}, Marc-Alexandre and Panov, Aleksandr I},

  journal={arXiv preprint arXiv:2206.00142},