Seonglae ChoScalable, Instructable, Multiworld Agent
arxiv.org
https://arxiv.org/pdf/2404.10179
Introducing SIMA, a Scalable Instructable Multiworld Agent
https://deepmind.google/discover/blog/sima-generalist-ai-agent-for-3d-virtual-environments/
SIMA 2
A general-purpose 3D agent that uses Gemini as its core reasoning engine, capable of goal inference, dialogue, self-explanation, and self-improvement beyond simple instruction following. It shows significantly improved generalization performance in unseen games and environments. It can adapt and act even in new 3D worlds generated by Genie Model.
It combines human demonstrations with self-directed learning to form a generational self-improvement loop: embodied agent + reasoning-heavy policy + imitation/self-play without world model.
SIMA 2: A Gemini-Powered AI Agent for 3D Virtual Worlds
Introducing SIMA 2, the next milestone in our research creating general and helpful AI agents. By integrating the advanced capabilities of our Gemini models, SIMA is evolving from an instruction-foll…
https://deepmind.google/blog/sima-2-an-agent-that-plays-reasons-and-learns-with-you-in-virtual-3d-worlds/
