DeepMind: SIMA 2: An agent that plays, reasons, and learns with you in virtual 3D worlds

Author: Google DeepMind - Bewertung: 46x - Views:620

We’re introducing SIMA 2, the next major milestone in general and helpful embodied AI agents. 👾



With Gemini integrated at its core, it moves beyond following basic instructions to think, learn, and collaborate in complex, 3D worlds.



🔵 Advanced reasoning: It can accomplish high-level goals in a wide array of games – describing its intentions, explaining what it sees, and outlining the steps it is taking.



🔵 Improved generalization: It can transfer concepts like “mining” in one game and apply it to “harvesting” in another - connecting the dots between similar tasks.



🔵 Self-improvement: Through trial-and-error and Gemini-based feedback, it can teach itself entirely new skills in unseen worlds without additional human input.



🔵 Adaptability: When tested in simulated 3D worlds created with our Genie 3 world model, it demonstrates unprecedented adaptability by navigating its surroundings, following instructions, and taking meaningful steps towards goals.



This research offers a strong path toward applications in robotics and another step towards AGI in the physical world.



Learn more at https://deepmind.google/sima

___



Subscribe to our channel https://www.youtube.com/@googledeepmind

Find us on X https://twitter.com/GoogleDeepMind

Follow us on Instagram https://instagram.com/googledeepmind

Add us on Linkedin https://www.linkedin.com/company/deepmind/