GROOT: Learning to Follow Instructions by Watching Gameplay Videos

We study the problem of building a controller that can follow open-ended instructions in open-world environments. We propose to follow reference videos as instructions, which offer expressive goal specifications while eliminating the need for expensive text-gameplay annotations. A new learning framework is derived to allow learning such instruction-following controllers from gameplay videos while producing a video instruction encoder that induces a structured goal space. We implement our agent G ROO T in a simple yet effective encoder-decoder architecture based on causal transformers. We evaluate G ROO T against open-world counterparts and human players on a proposed Minecraft SkillForge benchmark. The Elo ratings clearly show that G ROO T is closing the human-machine gap as well as exhibiting a 70% winning rate over the best generalist agent baseline. Qualitative analysis of the induced goal space further demonstrates some interesting emergent properties, including the goal composition and complex gameplay behavior synthesis. The project page is available at https: //craftjarvis-groot.github.io. Figure 1 | Through the cultivation of extensive gameplay videos, G ROO T has grown a rich set of skill fruits (number denotes success rate; skills shown above do not mean to be exhaustive; kudos to our artist Haowei).