Google's Robot AI Now Controls Humanoids Head to Toe

Robots just got a major upgrade that brings them closer to being genuinely helpful in our daily lives.

Google DeepMind announced Thursday that its Gemini Robotics 2 AI model can now control entire humanoid robots, coordinating movements from head to toe. The previous version could only manage upper body actions, but this new system enables robots to walk, crouch, stretch, and manipulate objects with precision.

Videos show Apptronik's Apollo 2 robot bending down to pick up a watering can and retrieving specific items from shelves. These might seem like simple tasks, but they require the kind of whole-body coordination that humans take for granted and robots have struggled to master.

The AI's improved dexterity stands out as particularly impressive. Robots equipped with five-fingered hands can now seal Ziploc bags, tie trash bags, and unscrew lightbulbs—tasks that demand the delicate touch we use dozens of times each day.

Google DeepMind also upgraded its Gemini Robotics ER model, which helps robots understand their surroundings and follow multi-step instructions. The new version better recognizes when tasks start and finish, allowing robots to work on projects over extended periods without losing track.

Even more exciting: different types of robots can now work together as a team. In one demonstration, Apollo 2 directs Google's dual-arm robot to organize tools while cleaning a garage, showing how diverse machines might collaborate in real workplaces.

The Bright Side

Safety improvements make this technology even more promising for real-world use. Google DeepMind calls this their safest robotics model yet, with enhanced ability to detect nearby humans and automatically stop if someone gets too close.

The company acknowledges their robots still need to improve movement speed. But this whole-body coordination represents a crucial step toward robots that can handle complex, practical tasks in homes, warehouses, and healthcare settings.

Google DeepMind has also enhanced its on-device model, which lets robots operate without internet connections. This means more reliable performance in areas with spotty connectivity and greater privacy for users who want robots handling tasks in their homes.

The progress signals a shift from robots as novelties to robots as practical helpers that could ease physical burdens and make daily life more manageable for people with mobility challenges or demanding jobs.