TLDR: Google DeepMind has introduced Gemini Robotics On-Device, a new generative AI model designed to run directly on robots, enabling them to perform complex tasks with low latency and without constant cloud connectivity. This advancement leverages a vision-language-action (VLA) model, allowing robots to understand multimodal commands and adapt to novel situations with remarkable dexterity, opening new avenues for automation across various industries.
Google DeepMind has announced a significant leap in robotic artificial intelligence with the introduction of Gemini Robotics On-Device, a specialized generative AI model engineered to operate directly on robotic hardware. This innovation aims to overcome traditional challenges of latency and connectivity by enabling robots to process commands and perform actions locally, eliminating the need for continuous cloud server interaction.
Building upon the foundational Gemini Robotics VLA (vision language action) model, Gemini Robotics On-Device is optimized for efficiency, requiring minimal computational resources while delivering robust performance. The model empowers robots with advanced capabilities, including the understanding of complex multi-modal queries that combine visual, audio, and text inputs. It allows robots to reason about physical spaces, generalize their understanding to novel situations—such as encountering new objects or environments—and respond dynamically to everyday commands, even adapting to sudden changes in instructions or surroundings without further human input.
Demonstrations have showcased the model’s impressive dexterity, with bi-arm robots performing intricate tasks like unzipping bags, folding clothes, zipping lunchboxes, drawing cards, and pouring salad dressing. The model’s adaptability is also a key highlight; developers can fine-tune it for new tasks with as few as 50 to 100 demonstrations, indicating its strong ability to generalize foundational knowledge.
Carolina Parada, Head of Robotics at Google DeepMind, expressed surprise at the model’s strength, stating, ‘The Gemini Robotics hybrid model is still more powerful, but we’re actually quite surprised at how strong this on-device model is.’ Google emphasized that ‘Gemini Robotics On-Device marks a step forward in making powerful robotics models more accessible and adaptable — and our on-device solution will help the robotics community tackle important latency and connectivity challenges.’
Also Read:
- Google’s Gemini AI Gains Default Access to Android Communications, Raising Privacy Concerns
- Google’s Advanced AI Search Mode Now Fully Available to All Users in India
Google is currently making the Gemini Robotics SDK (Software Development Kits) available to a select group of testers and companies through a trusted tester program, inviting feedback for further refinement. This on-device AI model holds immense potential for automating work across major industries, including consumer electronics, food, and automotive, where robots could assemble products, prepare goods, and even operate factories 24/7. Furthermore, its offline functionality opens up possibilities for critical applications in environments with limited connectivity, such as space exploration or disaster response, marking a new chapter for autonomous robot control and adaptability.
