Hustler Words – Google DeepMind unveiled Gemini Robotics On-Device, a groundbreaking language model enabling robots to operate autonomously, even without internet connectivity. Building upon the March release of Gemini Robotics, this new iteration empowers developers to control and refine robotic movements using simple natural language prompts, eliminating the need for constant cloud connection. This represents a significant leap forward in robotics, moving away from cloud dependency towards localized intelligence.
Benchmark tests reveal Gemini Robotics On-Device achieves performance comparable to its cloud-based counterpart. While Google refrained from specifying competitors, they assert superior performance compared to other on-device models in general benchmarks. Demonstrations showcased robots executing complex tasks such as unzipping bags and folding clothes, initially trained on ALOHA robots but successfully adapted to the bi-arm Franka FR3 and Apptronik’s Apollo humanoid robot. Remarkably, the Franka FR3 demonstrated adaptability, handling unseen scenarios and objects, including industrial assembly line tasks.
Further enhancing developer capabilities, Google DeepMind is releasing a Gemini Robotics SDK. This SDK allows developers to train robots on new tasks using the MuJoCo physics simulator, requiring only 50 to 100 demonstrations. This streamlined training process significantly reduces the time and resources needed to integrate new functionalities into robotic systems.
Related Post
Google’s advancement places them at the forefront of a burgeoning field. Competitors like Nvidia (building foundation models for humanoids), Hugging Face (developing open models, datasets, and even robots), and RLWRLD (creating foundational models for robots) are also making strides in the robotics AI space. However, Google’s Gemini Robotics On-Device stands out with its impressive on-device performance and ease of use, promising a new era of accessible and adaptable robotics.
Leave a Comment