This post was originally send out to my newsletter subscribers. The newsletter documents all my updates and imrpovements to the Lego Sorting bot. It is an almost monthly newsletter (in practice, I send out about 6 newsletters per year). You can read previous editions here, or you can subscribe using the form below.
Hello LEGO enthusiasts,
Hello and welcome to the Lego sorter newsletter. As a reminder, you are receiving this because you expressed interest in being kept up-to-date on my adventures in sorting Lego bricks with AI and various contraptions. If you no longer want to receive these mails, there is an unsubscribe link at the bottom.
This month, I’ve been making exciting progress with my machine, including a new integration with Brickognize that makes sorting bricks smarter and faster. If you’re curious about the latest updates and challenges, keep reading!
Integrating Brickognize
As suggested by you, the reader (Thanks!), I integrated Brickognize into the Lego sorting machine, which eliminates the need to collect training data manually and rely on a powerful GPU for training. With this update, you can start sorting bricks straight away using the brickognize branch of the software.
Currently, the integration is available in the pi/predict.py script, which offers two options for sorting: the old local AI method or the new Brickognize-based approach. You can select your preference in the pi/04_sorting_bricks.ipynb notebook.
While still in the early stages, I’m planning to create a standalone application for sorting that’s more user-friendly than Jupyter Notebook. Stay tuned for updates!
Failure modes
As I continue testing the machine, I’ve been documenting potential failure modes to refine its performance and reliability. Here are the main challenges I’ve identified so far:
- Feeding Issues:
- Bricks can get stuck in the initial feeder belt, which isn’t wrapped in paper.
- Larger bricks sometimes jam the middle conveyor belt, requiring manual intervention.
- Brick Placement:
- Two pieces can drop in front of the camera simultaneously, especially smaller parts, leading to misclassification.
- Some bricks are hard to distinguish from a top-down view. I’m exploring angled or multi-angle photos to improve accuracy.
Writing these down helps me brainstorm solutions and focus on the most impactful fixes. The feeder section, in particular, has been frustrating—time to revisit its original design!
What’s Next?
As we approach the end of 2024, I want to wish you an early Happy New Year! It’s unlikely I’ll send another newsletter before January, so here’s what’s next:
I’m planning to transition the machine to use Lego Powered Up motors and the Pi Build HAT, which will simplify the setup and significantly speed up operations. To prepare, I’ve already picked up the Audi RS Q e-tron (42160)—a fun build that also comes with three motors I need!
Thank you for following my journey. Your feedback and interest have been incredibly motivating, and I look forward to sharing more progress in 2025.
See you next year,
Peter