🆕 Introducing JAT, the first open-source multi-modal, multi-task multi-domain agent! 🤖 A step toward open generalist agents! 🚀 📰 Blog:

Huge kudos to @ClementRomac @edwardbeeching @Thom_Wolf for all the work you've done.

- 📄 Paper: - 💻 Code: - 🗂️ Dataset: - 🤖 Model:

Very cool! Did you consider using Go-Explore for the Atari trajectories? It solves all Atari games.

Go-Explore is definitely a golden choice for Atari. Having pre-trained Go-Explore agents and a trajectory dataset hosted of 🤗 Hub would allow JAT to be trained straight away with the Go-Explore data!

This video is so insane

Does JAT use the same network and same weight for all tasks (i.e. atari, mujuco, minigrid)? That's genuinely fascinating! I'm working on similar things recently, i.e. learning a world model to train multi-task RL agents, and your work looks really promising and inspiring 🤗

Yes, same weights for all tasks!

But all I want is a multi-fruit juice

@josh_bickett

Thanks for sharing and making aware of it! 👏👀