Apollo humanoid robot tackles unknown objects with Google DeepMind

Google DeepMind, in collaboration with Apptronik, has demonstrated significant advances in humanoid robotics through the Apollo robot, which can perform complex tasks involving unfamiliar objects by responding to natural-language instructions and adapting in real time. Powered by DeepMind’s Gemini 3 and Gemini Robotics AI models, Apollo showcases capabilities such as packing lunches, sorting laundry, and handling unpredictable household environments without needing retraining for each new task or setting. This progress reflects improvements in AI models and hardware, aiming toward versatile, general-purpose humanoid robots that can understand instructions, plan multi-step actions, and interact safely around people and fragile items. A notable innovation from DeepMind is the development of an “inner monologue” system, where robots internally describe observed human actions in natural language to better understand and remember tasks. This technique enables “zero-shot” learning, allowing robots to handle new objects and tasks without prior training, while also reducing the computational resources needed for training. Additionally, DeepMind introduced “Gemini Robotics On-