Google DeepMind has introduced Genie 4, a new AI world model that generates interactive environments in real time, alongside an experiment showing how the technology could support humanoid robotics research. In one demonstration, the company paired Genie 4 with Gemini 4 Argon to simulate Apptronik’s Apollo humanoid playing checkers, with the AI breaking the task into steps as the virtual scene unfolded.
The October 27 announcement marks an expansion of DeepMind’s work on world models, which attempt to predict how environments respond to actions. Unlike a conventional video generator that produces a fixed sequence, Genie 4 allows users to alter a scene while it is running. Natural-language instructions can change the weather, introduce objects or redirect the camera, while the model generates corresponding visuals and audio, including environmental sounds and dialogue.
DeepMind is also using Gemini models to make these generated worlds more responsive to complex tasks. Gemini 3.8 Flash can observe a simulation, plan a sequence of actions and assess whether individual steps appear to have succeeded. In the Apollo experiment, Gemini 4 Argon supplied task instructions to Genie 4 as the checkers scenario was rendered. The demonstration illustrates a possible route toward generating and adapting robot-training scenarios without having to stage every situation in the physical world.
The Apollo example took place entirely in simulation. DeepMind has not shown that the generated checkers sequence taught a physical humanoid a new skill or improved its real-world performance. For robot developers, the value of such systems will depend on whether simulated contacts, object movements and failures correspond reliably to what happens on actual hardware.
Genie 4 remains an early research model. DeepMind acknowledges that objects and visual details can drift during longer sessions and that spatial accuracy, complex physical interactions and long-term consistency still need improvement. Access is initially limited to trusted testers, with broader availability planned through Project Genie and early-access developer channels.
Watch here: https://www.youtube.com/watch?v=sFfPkUJd_Jk
