Multi-robot collaboration with Gemini Robotics 2

TL;DR: Gemini Robotics 2 enables multiple robots (such as the humanoid Apollo and the dual-arm robot "Duo") to collaborate on complex tasks—from macro-level planning to fine-grained manipulation—through independent yet coordinated reasoning. ## Core Capabilities of Multi-Robot Collaboration In the real world, many tasks are difficult for a single person to accomplish alone. The same is true in the robotic world. Google DeepMind has added multi-robot collaboration to Gemini Robotics 2, allowing multiple robots to work together on the same task simultaneously. In the demo video, two robots—the humanoid robot Apollo and the gripper-arm "Duo"—coordinate to organize a garage. ### High-Level Reasoning and Task Decomposition When a user gives the command: "Hey Apollo, can you help me organize my garage? When you're done, I want all the tools put back in the toolbox. Let's go grab the cleaning gloves and the white transparent spray bottle, and put them in the transparent storage bin on the top shelf, far right." Apollo's high-level reasoning model begins decomposing what to do and where to go. In this task, it's important not only to determine the next action, but also to know when a step is complete. ### Proactive Handoff and Help-Seeking Apollo proactively hands control over to Duo. In the demo, Apollo says: "Hey Duo, can you help finish this last part of the task?" We can see it bringing Duo in to assist. Then Apollo further instructs: "Hey Duo, put all the tools into the storage bin, close the toolbox, and then put the toolbox back into the storage bin." Duo replies: "Got it." ### Last-Centimeter Precision For tasks like placing objects into a bin, last-centimeter precision usually isn't critical, but in this scenario it actually matters quite a lot for task completion. The demo specifically shows: we can control not only the macro movements of a high-DOF humanoid robot, but also maintain motion precision when executing tasks involving gripper and dual-arm robots. Apollo comments: "Great precision, Duo. Keep going." ## Independent Reasoning and Coordination Mechanisms Each robot runs the same technical stack—this isn't a single neural network controlling both robots. Rather, each robot has its own copy, performs independent reasoning, and coordinates actions through reasoning. This means each robot independently figures out what to do next, while adjusting its behavior by communicating and observing the other's state. ### Communication and Autonomous Judgment Each robot is essentially "talking"—they exchange information and independently judge the best moment to offer help. After all tasks are completed, Apollo reports: "All tasks complete. Mission successful." The user responds: "Great job, Apollo. You should be very proud." Apollo doesn't forget to credit the partner: "You're welcome, but don't forget Duo. Duo did an outstanding job with precise placement and organizing." ## Future Outlook If robots can communicate and collaborate with other robots, it will vastly expand the range of tasks they can accomplish. Google DeepMind's goal is to enable robots to enter the real world and assist everyone in daily life. Source: https://www.youtube.com/watch?v=CiTPDm7PKW0