DeepMind Enables Apollo Humanoid To Manipulate Objects
Google DeepMind and Apptronik recently demonstrated the Apollo humanoid using Gemini 3 and Gemini Robotics models to manipulate unfamiliar, oddly shaped or soft objects and to follow verbal instructions in controlled lab demos. DeepMind also filed a patent for an "inner monologue" training technique and released a compact Gemini Robotics On-Device model, which together aim to enable zero-shot learning and more efficient on-robot autonomy.
Key Points
- 1Shows Apollo manipulating unfamiliar, oddly shaped and soft objects while following spoken instructions in lab demonstrations.
- 2Highlights inner monologue and on-device Gemini Robotics to improve data efficiency and enable zero-shot robotic learning.
- 3Enables practitioners to explore on-robot autonomy and zero-shot transfer for real-world humanoid deployment and adaptation.
Scoring Rationale
Demonstrations, patent and on-device model indicate substantial robotics progress, but results remain lab-limited and unproven in complex real-world settings.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
