Gilwoo Lee

CTO of Industrial Next.

[Gilwoo Lee @gilwoo_lee]

@rosario_scaling @TaraSadjadpour @SharpaRobotics oh this is super useful

[Lukas Ziegler @lukas_m_ziegler]

World action models solving cable insertion at scale! 📱

During @AutomateShow, Industrial Next showed a world action model application for actual manufacturing.

Cable insertion is brutal to automate. Cables break easily. They're never in the same position on the production line. Right now, this step is manual for virtually every phone ever made.

The challenge? High-precision, tactile work with deformable objects.

This company built a foundation model specifically for tiny electronics parts. They rejected third-party datasets. Quality matters more than quantity when precision is critical.

Two hours of extremely high-quality data beats months of mediocre data. They use teleop data combined with a corpus of basic manufacturing techniques to give the model context.

I had a chance to speak with Lukas Pankau and he explained me a bit of this and a bit of that.

For example that simulation doesn't help here. Deformable objects like wires are nearly impossible to simulate accurately. No physics simulator is good enough yet. So they collect real data, train on it, deploy it.

The model can handle long-horizon tasks in a single shot, maintaining sub-millimeter precision across an entire sequence without intermediate corrections.

They own the data collection, the model training, and the deployment. They position themselves as a "world platformer," not just a model vendor.

They do integration work because traditional system integrators still make money on software complexity, not on clean solutions.

Accoridng to them that's misaligned with where manufacturing is heading.

~~
♻️ Join the weekly robotics newsletter, and never miss any news → https://t.co/GoA3ZuwoPB

[Tom Silver @tomssilver]

This week's #PaperILike is "Bayesian Residual Policy Optimization" (Lee et al., 2020).

I like this POMDP approach because it reduces the problem to figuring out a good set of "clairvoyant experts".

PDF: https://t.co/Uv3QLLVlyE

[Gilwoo Lee @gilwoo_lee]

@tomssilver Thanks!! :)

[Yunchu Zhang @yunchuzh]

How should a robot perceive the world? What kind of visual representation leads to robust visuomotor policy learning for robotics?

Policies trained on raw images are often fragile—easily broken by lighting, clutter, or object variations—making it challenging to deploy policies learned via imitation learning in high variability test conditions. This same fragility is also reflected in the difficulty in transferring visuomotor policies from simulation to reality for robotic manipulation.

Introducing ATK https://t.co/W8fcMO6Xui: an automatic task-driven method for selecting flexible keypoint-based visual representations that enables robust, generalizable robotic manipulation with minimal human effort.(1/8)👇

[Gilwoo Lee @gilwoo_lee]

Honored to be selected for the inaugural Google’s AI Academy: American Infrastructure program!

[Google for Startups @GoogleStartups]

These 15 startups are tapping into AI’s potential to help governments and communities tackle critical challenges at scale.

Learn more about the companies selected for the inaugural AI Academy: American Infrastructure program: https://t.co/H1aGywDc1e

[Google for Startups @GoogleStartups]

These 15 startups are tapping into AI’s potential to help governments and communities tackle critical challenges at scale.

Learn more about the companies selected for the inaugural AI Academy: American Infrastructure program: https://t.co/H1aGywDc1e

[Vinod Khosla @vkhosla]

Bipedal robots have the capacity to transform every vertical from eldercare to factories & farms. Few are preparing for how this will radically change GDP, productivity, and human happiness. These robots could create enough value to support the people they replace.

[Gilwoo Lee @gilwoo_lee]

Today, I am very excited to share that we’ve acquired all of AppHarvest’s robotics IP and assets. This acquisition will let us further diversify our greenhouse automation across various crops and tasks with multiple patents, including those already developed by Zordi.

https://t.co/J3jHoiNbXV

Just in 1 month since acquisition, we were able to integrate the core gripper technology into our software stack and latest ML models, and successfully demonstrated cherry tomato picking at our partner facility.

What’s more exciting is that the technology was previously developed & led by @zordifarm CTO, @rrknopf Ryan Knopf, during his tenure at AppHarvest and RootAI. Over years of development, this unique design for harvesting individual delicate fruits has previously achieved incredible speeds of over 1000 grape tomato fruits harvested per hour. I am grateful that have inherited that team’s amazing achievement and now have the opportunity to continue its commercialization to new heights!

[Gilwoo Lee @gilwoo_lee]

@asta_li @paradigm @sequoia Congrats!!!!

[Gilwoo Lee @gilwoo_lee]

I had an opportunity to share our vision with the Verge. AI and robotics will play a critical role in sustainable farming -- not just growing sustainably, but also bringing farms closer to where produce gets consumed, and enabling that at a massive scale.
https://t.co/vYRNuTD7ad

[Gilwoo Lee @gilwoo_lee]

More announcements coming soon!

Thanks a lot @siddhss5 & @KanuGulati for your awesome advice and support through this journey.

[Zordi @zordifarm]

https://t.co/k4idySIsme

Excited to make our first announcement!

[Gilwoo Lee @gilwoo_lee]

First step to accurate yield prediction is to know when and where your flowers are coming first.

[Gilwoo Lee @gilwoo_lee]

I'm thrilled to announce that @RRKnopf is now the CTO @zordifarm ! Since he joined us, our robots have become more powerful; our AI has been making tailored recommendations per crop x region. We've been attracting top talent in robotics & AI. Excited for our journey ahead!

[Gilwoo Lee @gilwoo_lee]

@chiefaioffice ML