00Pro · lessons free to read

Use AI efficiently with ROS 2

Learn where AI speeds up ROS 2 work and where it will crash your robot, by fixing the code an LLM actually gets wrong.

Intermediate10-14 hoursUbuntu 24.04, ROS 2 Jazzy, rclpy/rclcpp, an LLM coding assistant + MCP

AI + ROS 2SYSTEM / FAULT / EVIDENCE

Outcome

Prompt, review, debug, and ship ROS 2 with an AI assistant safely: QoS, MCP agents, RAG grounding, real-time safety, and perception gating.

Build

What you will produce

5 skills

An AI-assisted ROS 2 node (business logic from the model, real-time and safety structure owned by you) repaired concept by concept.

Prompting for correct ROS 2
AI code review (QoS, lifecycle, version drift)
MCP + agentic robot control
AI-assisted debugging
Real-time safety + perception gating

Sample incident

The AI's publisher connects to nothing

3 nodes

Fix the QoS the LLM left off so the sensor publisher matches its subscriber.

01LLM-generated publisher→
02QoS profile→
03sensor subscriberOUT

Inspect sample case

Full syllabus

The eight modules

8 modules / 24 lessons

01
Prompting an LLM for correct ROS 2
An LLM writes plausible rclpy fast, but it drops the ROS 2 specifics (QoS, executors, lifecycle) that decide whether nodes actually talk.
Guided lab Independent lab
02
Generating nodes, launch, URDF, and messages
AI is fastest at boilerplate (launch files, URDF, .msg) but it hallucinates physics values and reaches for deprecated Gazebo Classic tags.
Guided lab Independent lab
03
MCP and agentic control of ROS
MCP servers let an LLM agent call ROS tools directly. That is dangerous when a tool publishes without validating what it sends.
Guided lab Independent lab
04
Debugging ROS with AI in the loop
AI reads logs and explains tf, QoS, and lifecycle errors well. You close the loop by verifying its fix against the running system.
Guided lab Independent lab
05
Grounding AI in the real ROS 2 docs
Un-grounded LLMs hallucinate APIs and mix Humble and Jazzy. Retrieval over the actual docs kills both.
Guided lab Independent lab
06
Real-time and safety: where AI can't help
LLMs write code that runs, not code that meets a deadline. In a control loop, that difference is a crash.
Guided lab Independent lab
07
Embodied AI: great planner, poor perceiver
LLMs and VLMs plan well but hallucinate perception. Close the loop with real sensors before you act.
Guided lab Independent lab
08
Ship an AI-assisted ROS system
Put it together: an AI-assisted node with a persisted checkpoint and a safe, idempotent recovery path.
Guided lab Independent lab

Playgrounds

Experiment in the browser

3 live

Small, functional experiments that reveal the math and evidence behind this course’s failures. Open them alongside the labs.

Live controls

LLM Prompt Token Budget

Add up system, history, and user tokens against the context window, leaving room for the reply, and see when a request overflows.

context window / token budget / LLM prompt

Live controls

Tool-Call Schema Validator

Validate a model's tool call against a JSON schema: required parameters, argument types, extra properties, and malformed JSON.

tool use / JSON schema / function calling

Live controls

RAG Chunk Retrieval Explorer

Tune chunk size, overlap, and top-k over a document and see how a boundary-spanning answer is lost when the overlap is too small.

RAG / chunking / retrieval recall

Loading system evidence

00Pro · lessons free to read

Use AI efficiently with ROS 2

Learn where AI speeds up ROS 2 work and where it will crash your robot, by fixing the code an LLM actually gets wrong.

Intermediate10-14 hoursUbuntu 24.04, ROS 2 Jazzy, rclpy/rclcpp, an LLM coding assistant + MCP

AI + ROS 2SYSTEM / FAULT / EVIDENCE

Outcome

Prompt, review, debug, and ship ROS 2 with an AI assistant safely: QoS, MCP agents, RAG grounding, real-time safety, and perception gating.

Build

What you will produce

5 skills

An AI-assisted ROS 2 node (business logic from the model, real-time and safety structure owned by you) repaired concept by concept.

Prompting for correct ROS 2
AI code review (QoS, lifecycle, version drift)
MCP + agentic robot control
AI-assisted debugging
Real-time safety + perception gating

Sample incident

The AI's publisher connects to nothing

3 nodes

Fix the QoS the LLM left off so the sensor publisher matches its subscriber.

01LLM-generated publisher→
02QoS profile→
03sensor subscriberOUT

Inspect sample case

Full syllabus

The eight modules

8 modules / 24 lessons

01
Prompting an LLM for correct ROS 2
An LLM writes plausible rclpy fast, but it drops the ROS 2 specifics (QoS, executors, lifecycle) that decide whether nodes actually talk.
Guided lab Independent lab
02
Generating nodes, launch, URDF, and messages
AI is fastest at boilerplate (launch files, URDF, .msg) but it hallucinates physics values and reaches for deprecated Gazebo Classic tags.
Guided lab Independent lab
03
MCP and agentic control of ROS
MCP servers let an LLM agent call ROS tools directly. That is dangerous when a tool publishes without validating what it sends.
Guided lab Independent lab
04
Debugging ROS with AI in the loop
AI reads logs and explains tf, QoS, and lifecycle errors well. You close the loop by verifying its fix against the running system.
Guided lab Independent lab
05
Grounding AI in the real ROS 2 docs
Un-grounded LLMs hallucinate APIs and mix Humble and Jazzy. Retrieval over the actual docs kills both.
Guided lab Independent lab
06
Real-time and safety: where AI can't help
LLMs write code that runs, not code that meets a deadline. In a control loop, that difference is a crash.
Guided lab Independent lab
07
Embodied AI: great planner, poor perceiver
LLMs and VLMs plan well but hallucinate perception. Close the loop with real sensors before you act.
Guided lab Independent lab
08
Ship an AI-assisted ROS system
Put it together: an AI-assisted node with a persisted checkpoint and a safe, idempotent recovery path.
Guided lab Independent lab

Playgrounds

Experiment in the browser

3 live

Small, functional experiments that reveal the math and evidence behind this course’s failures. Open them alongside the labs.

Live controls

LLM Prompt Token Budget

Add up system, history, and user tokens against the context window, leaving room for the reply, and see when a request overflows.

context window / token budget / LLM prompt

Live controls

Tool-Call Schema Validator

Validate a model's tool call against a JSON schema: required parameters, argument types, extra properties, and malformed JSON.

tool use / JSON schema / function calling

Live controls

RAG Chunk Retrieval Explorer

Tune chunk size, overlap, and top-k over a document and see how a boundary-spanning answer is lost when the overlap is too small.

RAG / chunking / retrieval recall

Use AI efficiently with ROS 2

Prompt, review, debug, and ship ROS 2 with an AI assistant safely: QoS, MCP agents, RAG grounding, real-time safety, and perception gating.

Prompting an LLM for correct ROS 2

Generating nodes, launch, URDF, and messages

MCP and agentic control of ROS

Debugging ROS with AI in the loop

Grounding AI in the real ROS 2 docs

Real-time and safety: where AI can't help

Embodied AI: great planner, poor perceiver

Ship an AI-assisted ROS system

LLM Prompt Token Budget

Tool-Call Schema Validator

RAG Chunk Retrieval Explorer

Use AI efficiently with ROS 2

Prompt, review, debug, and ship ROS 2 with an AI assistant safely: QoS, MCP agents, RAG grounding, real-time safety, and perception gating.

Prompting an LLM for correct ROS 2

Generating nodes, launch, URDF, and messages

MCP and agentic control of ROS

Debugging ROS with AI in the loop

Grounding AI in the real ROS 2 docs

Real-time and safety: where AI can't help

Embodied AI: great planner, poor perceiver

Ship an AI-assisted ROS system

LLM Prompt Token Budget

Tool-Call Schema Validator

RAG Chunk Retrieval Explorer