Google's Gemini Bots Master Real-World Tasks

Google DeepMind has announced a significant advancement in robotics with its Gemini Robotics 2 platform, which leverages a new version of the Gemini AI model to control a variety of robots. Demonstrations, captured in real-time footage, show humanoid robots performing intricate tasks autonomously, including putting tape into a boombox, screwing in a lightbulb, and tying garbage bags. This new system offers "intelligent whole-body control," a step beyond previous robotic arms performing simpler actions. The AI's multimodal understanding is achieved by integrating several specialized AI models, including vision language models. While impressive, these robots are currently trained for specific tasks seen in the video, rather than being general-purpose. Google DeepMind views this as progress toward "physical AGI," aiming for robots capable of any human-like action. The company is prioritizing safety, implementing multi-layered guardrails and a new benchmark, ASIMOV-Agentic, to detect potentially harmful outcomes, acknowledging the increased risks associated with AI-powered physical robots. Despite rapid advancements, these robots are in early development stages, with no immediate consumer release planned, contrasting with more aggressive timelines from competitors.