Dimensional Framework Adds Local AI
Dimensional, an open-source framework for programming robots, has added support for running AI models locally on-device. The new capability was demonstrated on an NVIDIA Jetson Thor using the Nemotron model, enabling physical agents to operate with more control and potentially lower latency.
- NVIDIA's Jetson Thor is a high-performance computer designed for AI and robotics, featuring a Blackwell GPU that delivers up to 2070 TFLOPS of performance for AI tasks. It includes a 14-core ARM Neoverse CPU and 128GB of LDDDR5X memory, providing server-class performance in a compact module for edge deployments. - The Nemotron models are a family of open, scalable language models from NVIDIA, designed for a range of applications including agentic AI. They are optimized for high-throughput and efficiency, making them suitable for deployment on edge devices like the Jetson Thor. - Running large AI models like Nemotron directly on a robot's hardware eliminates the need for cloud connectivity, which can significantly reduce latency in decision-making and physical actions. This is critical for autonomous systems that need to react in real-time to dynamic environments. - On-device AI processing enhances data privacy and security by keeping sensitive information, such as data from cameras and other sensors, on the local device instead of transmitting it to the cloud. - This shift towards powerful on-device AI, often referred to as "Physical AI," is a major trend in the robotics industry, with the market for embodied AI projected to grow from \\$4.44 billion in 2025 to \\$23.06 billion by 2030. - For embedded systems roles, this trend highlights the growing importance of skills in optimizing AI models for performance on specialized hardware and experience with platforms like NVIDIA's Isaac for robotic AI software development. - The open-source nature of frameworks and models like Nemotron allows for greater customization and transparency in developing robotics applications, a key consideration for both commercial and research projects. - Other companies are also heavily invested in on-device AI for robotics; for example, Google DeepMind has introduced Gemini Robotics On-Device, a vision-language-action (VLA) model optimized to run locally on robotic hardware.
