All articles tagged with #gemini robotics
Google DeepMind's Gemini Robotics 2 fuses a vision-language model with two vision-language action models to let humanoid and other robots perceive scenes and perform tasks, as shown in demos with Apptronik Apollo 2. While it signals progress toward physical AGI, it also highlights safety risks and introduces guardrails and the ASIMOV-Agentic benchmark to measure potentially harmful or uncertain outcomes.
Google DeepMind’s Gemini Robotics 2 expands humanoid-robot control to full-body motion—feet through fingertips—enabling walking, crouching and dexterous five-finger hand tasks, as demonstrated on Apptronik’s Apollo 2. The update also introduces Gemini Robotics ER 2 for extended, safer multi-step tasks and an on-device model that runs offline and adapts to new robot designs, with demos showing multi-robot collaboration and safer operation around humans.
Google DeepMind has introduced Gemini Robotics On-Device, a new AI model that enables robots to perform tasks locally without internet, demonstrating capabilities like unzipping and folding, and adaptable to various robot types, with a new SDK for developers.