Intelligent whole-body control with Gemini Robotics 2
A Google DeepMind presenter explains how Gemini Robotics 2 combines reasoning and whole-body control to help Apollo handle coordinated tasks in a cluttered environment.
In this talk
- 0:04
Why whole-body control is difficult
People coordinate many muscles and joints while moving through spaces built for human bodies. Robots need to coordinate their joints and actuators, keep their balance, and react quickly when their posture puts them at risk of tipping.
- 0:40
Reasoning and action work together
Gemini Robotics interprets the scene and natural-language instructions, then calls a vision-language-action model to generate robot movements. For packing sports gear, Apollo uses calendar details to identify Jesse’s pickleball match and Jeremy’s baseball game.
- 1:16
Packing in a cluttered space
The robot must find the requested object and place it correctly, while coordinating its body through a cluttered environment. The presenter says it can recognize some failures and try again; Apollo retrieves Jesse’s paddle from a middle shelf and reports that packing is complete.
- 1:57
Reacting to a new challenge
The presenter asks Apollo to locate a tote bag placed on the floor to its left, pick it up, and put it on a table. This is framed as a test of the robot’s ability to react to a new instruction and object placement.
- 2:12
A path to useful robots
The presenter describes the goal as a generalist robot that can perform more useful tasks in real-world settings. Whole-body control is presented as necessary for that goal, with AI described as a missing piece in robotics.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.