Google DeepMind stellt zwei neue KI-Modelle vor, die Roboter intelligenter und vielseitiger machen.
In Kürze
- Gemini Robotics 1.5 trifft eigenständige Entscheidungen
- Kombination von Vision- und Sprachmodellen für komplexe Aufgaben
- Sicherheitsaspekte werden eigenständig berücksichtigt
Google DeepMind’s New AI Models for Robotics
Google DeepMind has recently introduced two new AI models for robotics: Gemini Robotics 1.5 and Gemini Robotics-ER 1.5. These models bring a fresh approach to the world of artificial intelligence by adopting an „agentic“ approach. This means that robots are no longer just executing commands but can also make independent decisions and plans.
Integration of Vision and Language Models
A central aspect of these models is the combination of vision and language models. As a result, robots are capable of translating visual information and instructions into movements. They can handle complex, multi-step tasks, significantly expanding their potential applications.
Features of Gemini Robotics 1.5
Particularly impressive is Gemini Robotics 1.5. This model can reflect on its own actions, providing greater transparency in the decision-making process. Additionally, it enables the transfer of actions and movements between different types of robots. This not only accelerates learning processes but also facilitates adaptation to new environments.
Capabilities of Gemini Robotics-ER 1.5
Gemini Robotics-ER 1.5, on the other hand, excels in planning and logical thinking within physical spaces. It utilizes online content to optimize task solutions, achieving outstanding results in numerous academic benchmarks.
Google DeepMind’s Vision for AI-Driven Robots
With these developments, Google DeepMind aims to create AI-driven robots that are versatile in everyday use and can safely interact with humans. Safety is of utmost priority: the models are designed to autonomously consider safety aspects and act accordingly before becoming active.
These advances represent a significant step towards artificial intelligence that can also operate in the physical world.
Quellen
- Quelle: Google DeepMind
- Der ursprüngliche Artikel wurde hier veröffentlicht
- Dieser Artikel wurde im Podcast KI-Briefing-Daily behandelt. Die Folge kannst du hier anhören.
