Codestral ROS2 Generator tech description.

👋 I've just published a detailed technical overview in Russian of my Codestral ROS2 Generator project - an AI-powered solution that automatically generates ROS2 components (nodes, services, actions) using the Codestral model.

💡The article covers:
- Complete project architecture.
- Key classes and their interactions.
- The generation workflow from prompt to tested code.
- Performance metrics and evaluation.

Check out the full technical breakdown on my blog: Let's Go Design

#CodeGenerating #ROS #Codestral

We'll see...

Forecasts are getting more and more optimistic 🤔

Is Humanity Getting Dumber? @Science Says Yes

A recent study confirms a concerning trend: global intelligence levels have been declining since 2010. Researchers have found that many people struggle to focus on a single task, lose logical thinking skills, and spend excessive time on their smartphones.

Here are some key findings:

🔢 25% of adults in developed countries cannot solve basic math problems, with this figure rising to 35% in the U.S.
🧑‍🎓 Since 2015, there has been a sharp increase in 18-year-olds struggling to concentrate on a single topic.
📖 In 2022, half of those surveyed read no more than one book, while 45% of teenagers did not read at all.

The culprit? The way we consume information has changed. Social media, with its endless feeds and algorithm-driven content, overloads the brain, making it harder to seek out and process valuable information. Instead, people get stuck in an endless loop of low-value content consumption.

💡 The good news?
The brain remains highly adaptable—with effort, it’s possible to regain focus and sharpen critical thinking skills.

Time to take control of our attention!

I've been disappointed numerous times with highly-ranked models like Qwen that excel in benchmarks but underperform on real tasks. These benchmarks have started to feel like marketing tools—ways to generate hype with impressive tables and graphs rather than demonstrate genuine capability.What matters is genuine expertise. Human experts have credentials—education, years of field experience, common sense—and accountability for their decisions. But what about AI models? Sure, they can assist, but can we really trust them with complex real-world problems without human oversight? Will we ever truly believe in their "expertise"?My experience so far suggests caution—don't rely too heavily on AI advice. I'm curious to see how this evolves over time. 🤔

🌐 Codestral ROS2 Gen: Network Scanner Extension Now Available!

⚡️Here it is - the second release of Codestral ROS2 Gen with a powerful new feature: the NetworkScanner!

🔍 What's New in This Release
The NetworkScanner module with efficient network discovery capabilities:
• Fast ICMP-based scanning for host discovery
• Asynchronous packet handling for optimized performance
• Configurable scan parameters (timeout, interval, etc.)
• Seamless integration with the existing ROS2 generation pipeline

🛠 How It Works
The NetworkScanner uses optimized ICMP echo requests (pings) to detect active hosts on a network. It employs an innovative approach with:
• Synchronous packet sending for precise timing control
• Asynchronous response collection for efficient handling
• Smart timeout management for reliable results
• Clean ROS2 message publishing for network status information

🧪 Key Components
• network_scanner.py: Core scanning orchestration
• network_host.py: Host state management
• scan_operation.py: Context-managed scanning operations
• network_parser.py: Network targets parsing

Full codebase documentation is available on projects's GitHub pages 📙

🚀 Try It Yourself
The detailed example documentation shows how to generate your own network scanner nodes. You can even use the standalone nscan command-line tool for quick testing!

This extension builds upon core generation system, demonstrating how the Codestral generator can create complex, functional ROS2 components with system-level interactions. 🤖

#ROS2 #AI #NetworkScanning #Robotics #CodeGenerating #Codestral

🤩 Just tested Qwen 2.5-Omni and wow - it's impressive!

This free model accepts audio and video alongside text and images, plus it responds with both text and voice. We're just one step away from truly real-time AI conversation now - only need to cut down that input/response delay. Worth checking out here.

#QwenAI #MultimodalAI

👨‍🔬Testing Results: ROS2 Network Scanner Generation

I want to share the results from my test of ROS2 Network Scanner generation example.

After running 30 iterations of generating the ROS2 Network Scanner:
• Total test duration: ~6 hours 15 minutes
• Average successful generation time: ~2 minutes per attempt
• Distribution of attempts: Right-skewed (median: 4, mean: 6.7)

This means that, on average, the generator produces working code in about 13 minutes - quite reasonable performance for automated code generation in my opinion!

Failure Analysis
Looking at where generation stopped, the distribution clearly demonstrates the generator's stability:
• Over 80% stopped at the testing stage
• ~15% were successful attempts
• Only about 5% failed during the PARSING or GENERATION stages

Test Coverage Patterns
Examining the test pass rates revealed two distinct patterns:
• Basic functionality (7 tests): Node startup with valid/invalid parameters and handling overlapping scans using nscan utility
• Advanced scenarios (9 tests): Including handling invalid JSON format in the node <-> nscan interface and managing outdated scan results

This suggests that generating code with specific behavior for edge cases remains challenging.

I've included all metrics and analysis notebooks in my project repository, so feel free to explore the data yourself!

#ROS2 #AI #NetworkScanning #Robotics #CodeGenerating #Codestral #testing

💡Using AI for Coding - Part 1: Choose the White or Black Box

For maximum effectiveness when using AI to develop code, I recommend one of two opposite approaches. In my experience, these contrasting methods yield better results than any compromise solution.

⬜️ 1. The White Box: Pair Programming
In this approach, you effectively take on both 'driver' and 'navigator' roles simultaneously. The key principle is maintaining full control and visibility over all code because, ultimately, it's yours.
AI serves as an extremely helpful partner for discussing specific aspects of code architecture, design patterns, optimization techniques, and similar topics. However, never reduce yourself to a simple 'driver' who only performs copy-paste operations. This is a dead end that will quickly turn your codebase into a total mess!

⬛️ 2. The Black Box: Test-Driven Development
This is the approach I'm currently experimenting with in my AI code generator project. With this method, you might not even look at the final code you're developing at all. Instead, your main focus shifts to creating:
• Proper prompts
• Appropriate AI-model settings
• Comprehensive test suites

These elements together ensure your code works as expected, without you needing to understand every implementation detail.

Why Extremes Work Better
So we have two distinct cases: the white box with full control over code development, or the black box where you control only inputs and outputs. My experience suggests that any kind of "gray box" approach will be less efficient, primarily impacting your development skills and time investment.

Adopting a mixed "gray box" approach often gives you the worst of both worlds. Instead of boosting productivity, this middle ground typically creates unnecessary complexity and duplicates work without delivering the real benefits of either pure approach. You'll find yourself juggling opposing strategies rather than fully leveraging the strengths of either method.

What are your thoughts on these approaches? I'd be very interested in your comments 😁

#Thoughts #Experience