Boston Dynamics has just announced the integration of Google DeepMind's Gemini Robotics-ER 1.6 artificial intelligence model into its quadruped robot, Spot, allowing it to reason and handle complex situations autonomously. According to IEEE Spectrum, this is a step toward commercializing embodied AI in factory patrols and safety inspection tasks.
Key Developments
With DeepMind's new model, Spot can automatically detect hazardous debris, oil spills, or read pressure gauge faces without requiring detailed manual programming. The model helps the robot connect natural language to real-world actions. However, a major hurdle at present is that the model is entirely vision-only due to a lack of tactile sensor data for AI training.
To ensure safety, the system uses Google DeepMind's ASIMOV evaluation benchmark to filter out hazardous behaviors. A representative from Boston Dynamics stated they are conducting limited testing with a small group of customers before a broader rollout, aiming for over 80% reliability.
Why It Matters
This partnership opens up a practical path forward for the era of thinking robots. Instead of expecting an all-purpose robot to do household chores immediately, optimizing for specialized tasks like industrial infrastructure inspections serves as a necessary stepping stone. However, the challenge of processing physical real-world data remains substantial before robots can operate flawlessly in every situation.