Robots with AI Agents + Copilot
Griffin Norris demonstrates how AI agents and Microsoft Copilot can speed up robotics development by generating robot control code from natural language, connecting robots to AI models via MCP servers, and adding vision-based scene understanding for interactive, feedback-driven control.
Overview
This episode of Cozy AI Kitchen focuses on hands-on robotics development using AI agents and Microsoft Copilot to reduce the amount of “from scratch” robot programming.
Key themes covered in the session:
- Using natural language prompts to generate robot control scripts.
- Using MCP servers (Model Context Protocol) so AI agents can access robot APIs.
- Connecting robots to cloud-based AI models (including Azure AI) for reasoning and assistance.
- Adding vision AI for scene understanding (for example, identifying objects in the environment).
- Building reusable robotic “skills” instead of repeatedly generating raw control code.
- Running interactive control loops where the robot acts, observes results, and iterates with human oversight.
Chapters (from the video description)
- 00:01 — Robots + AI: The Big Challenge
- 00:15 — Meet Griffin Norris (General Robotics)
- 01:25 — Using Copilot + MCP for Code Generation
- 01:59 — From Docs to Working Robot Code
- 05:00 — Live Demo: Robot Stands and Walks
- 05:49 — Adding Vision AI + Scene Understanding
- 08:20 — Multi-Agent Systems & Faster Development
- 09:06 — Prebuilt Skills vs Raw Code Generation
- 11:05 — Controlling Robot with Natural Language
- 12:12 — Demo: Finding a Toaster with Vision AI
- 13:52 — Feedback Loops and Iteration
- 14:55 — Why Now Is the Best Time for Robotics
Tools and concepts mentioned
- AI agents and MCP (Model Context Protocol)
- Robot APIs and abstraction layers
- Vision AI / scene description models
- Multi-agent systems
- Simulation environments: AirSim and Isaac (as listed in the description)
- Human-in-the-loop safety for robotics
Speakers
- Griffin Norris (General Robotics)
- John Maeda (Host, Cozy AI Kitchen; Microsoft)