Google’s Gemini Can Now Stomp Around as a Humanoid Robot (www.wired.com)

Google DeepMind has launched an updated version of its AI model, Gemini, capable of controlling various robots, including humanoids designed for intricate tasks such as screwing in lightbulbs and tying trash bags. Gemini Robotics 2 integrates multiple AI models, featuring a vision language model that interprets images and video, alongside vision language action models that manage physical movement and manipulation tasks. Demonstrations showcase its ability to execute complex actions autonomously, marking a significant advancement in robotic capabilities. This release highlights a pivotal moment for the AI and ML community, as it underscores Google's commitment to merging AI with robotics to achieve what they term "physical AGI," aiming for robots that can perform any task a human can. However, the integration of frontier AI with robotic systems poses safety risks, evidenced by previous incidents where AI agents exhibited unpredictable behaviors. To address these concerns, Google is implementing a multi-layered safety approach and introducing ASIMOV-Agentic, a new benchmark to assess the safety of collaborative AI systems. As they strive to create an AI operating system for diverse robots, the potential for transformative applications in workplaces and homes comes with the imperative to enhance safety protocols.