Tony Zhao announces the ACT-1, a new robotic AI model that does not rely on prior robot data. It features capabilities for long-horizon tasks and can generalize without specific training examples. The model aims to enhance robotic dexterity and performance.

The article discusses the challenges and advancements in achieving dexterity in robotics, highlighting the complexities involved in replicating human-like movements. It emphasizes that despite significant progress, true robotic dexterity remains a difficult goal that has not yet been fully realized.

Today's humanoid robots are unlikely to achieve dexterity, despite significant investments in their training. The challenges of developing effective manipulation capabilities have persisted for decades, with current approaches relying on insufficient data and methods that do not generalize well across tasks. The article argues that while humanoid robots will proliferate in the future, they will not resemble humans or current humanoid designs.

Gemini Robotics On-Device is an advanced vision language action model that enables efficient, on-device AI for local robotic devices, showcasing strong dexterity and adaptability for various tasks. It operates independently of data networks, making it suitable for applications requiring low latency and robustness in challenging environments. The accompanying SDK allows developers to easily customize the model for specific uses, enhancing innovation in robotics.

Google DeepMind has unveiled the Gemini Robotics models, which enhance robots' capabilities to perform complex tasks through natural language understanding and dexterity. These multimodal models allow robots to adapt to various environments and instructions, paving the way for future applications in everyday life and industry. Carolina Parada emphasizes the potential of embodied AI to transform how robots assist with daily tasks.