Built AI agents for turtlesim and TurtleBot3 using LangChain – seeking feedback on LangGraph and MCP for robotics

Projects
,
1

Hi everyone,

I’ve recently been working on AI agent systems for controlling robots in ROS 2 environments, using TurtleSim and TurtleBot3. I implemented these agents using LangChain, and I’m now wondering if LangGraph might be a better fit for robotics applications, especially as the complexity of decision-making increases.

Here are the GitHub repos:

turtlesim agent: GitHub - Yutarop/turtlesim_agent: Draw with AI in ROS2 TurtleSim

turtlebot3 agent: GitHub - Yutarop/turtlebot3_agent: Control TurtleBot3 with natural language using LLMs

Now, I’d love your insights on a couple of things:

Would LangGraph be better suited for more complex, stateful behavior in robotic agents compared to LangChain’s standard agent framework?

Has anyone experimented with MCP (Model Context Protocol) in robotics applications? Does it align well with the needs of real-world robotic systems?

Any feedback, ideas, or relevant papers are greatly appreciated. Happy to connect if you’re working on anything similar!

2 Likes
2

I haven’t gotten to check out your work and no thoughts on langchain vs langgraph, but…

I saw a blender mcp that possibly could be good for synthesizing simulation environments. (Don’t have the link right now but let me know and I can find it)

And also, i’d love to have a gazebo mcp. Maybe this can be initially as simple as rendering videos from certain angles to be fed to LLMs as how the code is doing, and later on more control to make changes in the simulation as well.