Robotics Framework 'Dimensional' Adds Local Model Support
The open-source robotics programming framework Dimensional has added support for running AI models locally on-device. A developer demonstrated the framework running on an NVIDIA Jetson Thor with the Nemotron model as its brain, controlling agents in physical space. This feature allows for greater control and lower latency by reducing reliance on cloud-based processing for robot actions.
- The NVIDIA Jetson Thor is a high-performance computer for robots and other edge AI applications, featuring a Blackwell architecture GPU that can deliver up to 2,070 TFLOPS of performance for AI tasks. It is designed to run multiple generative AI models locally, allowing robots to perform complex tasks without constant reliance on the cloud. Agility Robotics is one of the companies exploring the use of Jetson Thor to run more powerful AI and reasoning models directly on their Digit humanoid robots. - NVIDIA's Nemotron model family is a series of open-source language models optimized for "agentic AI," where an AI can reason and carry out multi-step tasks. The "Nano" version of Nemotron is specifically designed for on-device intelligence, offering a balance of performance and efficiency that makes it suitable for deployment in robotics and IoT applications on hardware like the Jetson series. - The move toward on-device AI in robotics is driven by the need for lower latency, increased data privacy, and greater operational reliability in environments with intermittent or no internet connectivity. Frameworks and platforms like Ollama are making it easier for developers to run large language models locally on their own hardware. - A company named Trans-Dimensional Intelligence recently open-sourced a toolchain called EmbodiChain, which aligns with the trend of on-device AI for robotics. EmbodiChain focuses on training vision-language-action (VLA) models using 100% synthetic data and deploying them on real robots, aiming to accelerate the development of embodied intelligence. - This trend of "physical AI" represents a significant shift from AI that only processes digital information to AI that can perceive, reason, and interact with the 3D world. NVIDIA is building a comprehensive ecosystem to support this, including AI models like Nemotron, simulation platforms like Isaac Sim, and deployment hardware such as the Jetson Thor, to create a more streamlined pipeline for developing and deploying intelligent robots.
