Vibe Coding Robotics Part 10 Added a secondary camera focused on the gripper Next : generating CAD scripts to explore more advanced gripper mechanisms Code : Gemini 3.1 Pro Design : Nano Banana

There are a few gripper mechanisms I want to experiment with, but they're getting too complex for three.js primitives. So I'm shifting approach: generate CAD scripts first → create the models → import them into the three.js scene.

I've been experimenting with both online and offline CAD tools, and so far FreeCAD and OpenSCAD seem the most suitable for AI-generated scripts. Still figuring out the best way to prompt them. Claude has been giving the most consistent results so far, but there’s definitely room for improvement.

Ten parts in and you are literally vibe coding a robot into existence which is the most futuristic sentence that also happens to be completely real.

Is this compatible with SDF?

not yet

so good part 11: boston dynamics acquires Dilum

thanks! haha, I bet they are looking for experienced computer vision engineers too😉

@andrewarruda @llmgram

The jump from three.js primitives to AI-generated CAD scripts is such a smart progression — OpenSCAD's parametric approach should pair really well with LLM-generated geometry for rapid gripper iteration.

Love the DORFic aesthetic

@lio3D 😃

the secondary camera is a smart move. physical systems need tighter feedback loops than software does. a gripper failing silently because the visual check is ambiguous can corrupt your whole dataset before you notice. curious how you're handling edge cases where the camera can't distinguish a clean grab from a near-miss.

oh that UI is inspiring

Thanks, I was also trying to create some futuristic UIs related to things like protein folding. I'll post about some of those in the future.

soo cool

This is utterly fascinating! How close were you to producing something like this pre-AI? Did AI close a huge gap, small gap, or no gap but it lets you move faster?

The CAD script generation approach is the right call — parametric gripper geometry is way too complex to hand-code in primitives. Curious if you're planning to validate the generated scripts in simulation before importing, or just iterate visually in the scene.

cool addition with the secondary camera. CAD scripts will really open up new ideas for the gripper design.

wire this with ros2

In this series, I'll explore what LLMs can create without using something like ROS, and maybe start a separate series to compare them

Why not use @nvidia isaac sim or gazebo if you already have the CAD ?

CAD first workflow fits the pace of hardware prototyping. curious how you validate generated scripts before bringing them into play

awesome project

i think it is fantastic for ones that what to learn robotic without having the $$$

this is where vibe coding gets wild. software bugs are annoying, hardware bugs are expensive. curious how you're validating the gripper commands before they run on the real arm. part 10 already, the iteration speed is insane.

👀👀👀 🚀🚀🚀

where is other's part?

open source?

vibe-coding robotics with gemini generating cad scripts is the logical endpoint. > using ai to iterate physical mechanisms before a single part is printed > this changes prototyping from weeks to coffee breaks we're not just coding software anymore, we're compiling OBJECTS.

Vibe coding robotics with AI-controlled gripper cameras is genuinely next-level 🤖 the vision-to-action loop in physical hardware is so much harder than pure software. What's the latency like between the camera feed and gripper response?

Share script?

big deal

this looks so slick and clean, i like the layout a lot. the secondary camera view is a neat touch too.

super cool

Would you be interested in creating a whole environment like this for a robotics welding cell in the future? First digital and then it’ll be built in a shop for production

Vibe coding robotics is where this whole movement proves its value beyond web apps. CAD script generation from AI is going to collapse prototyping timelines for hardware teams. Curious how the gripper accuracy compares to hand-tuned kinematics at this stage.

Why does robotics have no documentation tools or VCS? Coming from software where is my Git!🤣

Nice touch with the gripper cam! Can't wait to see what crazy gripper designs you come up with ne...

Impressive CAD-to-sim workflow using Gemini. I'm watching Canadian robotics automate itself into obsolescence while UAE actually funds this caliber of technical depth.

and companies will raised 100 gazillion dollars to recreate a digital twin environment smh