Google DeepMind announced Thursday that its latest Gemini Robotics AI model can now control entire humanoid robots using whole-body motions [1].
This development marks a shift toward more versatile robotics. By coordinating the entire physical frame rather than isolated segments, AI can enable robots to navigate and interact with human environments more naturally.
The new system, Gemini Robotics 2, supports movements ranging from the robot's feet to its fingertips [1]. This is a significant expansion of capability, as the previous model focused primarily on controlling a humanoid robot's upper body [1].
DeepMind said the model allows robots to perform a wider range of actions [2, 3]. These tasks include cleaning up trash, and picking up watering cans [2, 3]. Such actions require a level of balance and coordination that involves the entire body to maintain stability while manipulating objects.
According to reports from The Verge, the model essentially acts as an AI brain for full-body humanoid control [1, 3]. This integration allows for a more seamless transition between locomotion and manipulation—the ability to move toward an object and interact with it in one fluid motion.
The ability to manage whole-body motions is critical for the deployment of humanoid robots in unstructured environments. While previous iterations were limited to torso and arm movements, the inclusion of lower-body coordination allows for a more comprehensive range of utility in domestic or industrial settings [1, 2].
“The latest version of its Gemini Robotics AI model can 'control entire humanoid robots.'”
The transition from upper-body control to whole-body coordination represents a critical hurdle in robotics. By integrating balance and locomotion with fine motor skills, Google DeepMind is moving humanoid AI closer to general-purpose utility, where robots can operate in human spaces without needing specialized tracks or fixed bases.
