The demonstration asks an Apollo humanoid and a smaller two-arm robot to tidy a garage together. The high-level model breaks the request into actions, locations and completion checks rather than treating the instruction as one indivisible sequence.
Apollo handles broad movement and calls the second robot when the task reaches a stage that needs more precise manipulation. The smaller system kits the tools, closes the container and returns it to the bin while both robots communicate about responsibility and completion.
Each robot runs its own copy of the reasoning and control stack. Their collaboration is orchestrated through language and individual decisions, rather than one neural network directly controlling both machines, which could let teams cover a wider range of real-world tasks.
Watch on YouTube
