AI BriefingNVIDIA Releases Guide for Building Robot Agents
Key point
NVIDIA has revealed how to build AI agents that operate in physical environments using DGX Spark and Reachy Mini.
Details
At CES 2026, NVIDIA announced new open models to expand the AI agent ecosystem. The core offerings are NVIDIA Nemotron for reasoning, NVIDIA Isaac GR00T N1.6, a VLA (Vision-Language-Action) model, and the NVIDIA Cosmos world foundation model.
In particular, NVIDIA presented a concrete methodology for combining a DGX Spark workstation with a Reachy Mini robot to build a personal AI agent that interacts with users in real time. This demo utilizes the following tech stack:
- Reasoning: NVIDIA Nemotron 3 Nano
- Vision: NVIDIA Nemotron Nano 2 VL
- Voice: ElevenLabs
- Framework: NVIDIA NeMo Agent Toolkit
This system adopts an open source approach that lets users directly control the model, prompts, tools, and the robot's actions. Developers can build their own physical AI agents using a local GPU environment or cloud infrastructure.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.